jaccen/Awesome-Gaussian-Skills

本技能是 [Awesome-Gaussian-Skills](https://github.com/jaccen/Awesome-Gaussian-Skills) 工具包的总入口与调度器。当前基线（**2026-10-21 快照**，`data/methods.json` 的 `generated_at`）：

| 单一事实源（方法库） | 仓库内 `data/methods.json`（改任何计数前先看它） | | 类目清单 | 仓库内 `data/categories.json`（26 个 canonical 类目 + 各类计数） | | GitHub 仓库 | https://github.com/jaccen/Awesome-Gaussian-Skills |

| 类目清单 | 仓库内 `data/categories.json`（26 个 canonical 类目 + 各类计数） | | GitHub 仓库 | https://github.com/jaccen/Awesome-Gaussian-Skills | | 在线方法检索器 | https://jaccen.github.io/Awesome-Gaussian-Skills/ |

| GitHub 仓库 | https://github.com/jaccen/Awesome-Gaussian-Skills |

| 在线方法检索器 | https://jaccen.github.io/Awesome-Gaussian-Skills/ |

| 在线书籍（空间与具身智能） | https://jaccen.github.io/Awesome-Gaussian-Skills/spatial-embodied-intelligence.html |

1. **arXiv API 编号查询**：确认 ID 可达且标题/摘要含方法名（`https://arxiv.org/abs/<id>`）

- **2D GauSS-MI** [arXiv:2609.21516](https://arxiv.org/abs/2609.21516) (arXiv 2026) — Efficient active scene reconstruction on 2DGS with online mapping and a probabilistic reliability model for view selection.

- **2D GauSS-MI** [arXiv:2609.21516](https://arxiv.org/abs/2609.21516) (arXiv 2026) — Efficient active scene reconstruction on 2DGS with online mapping and a probabilistic reliability model for view selection.

- **2D-SuGaR** [arXiv:2605.00569](https://arxiv.org/abs/2605.00569) (Eurographics 2026) — 2DGS enhanced with monocular depth/normal priors

- **2D-SuGaR** [arXiv:2605.00569](https://arxiv.org/abs/2605.00569) (Eurographics 2026) — 2DGS enhanced with monocular depth/normal priors

- **3DSS** [arXiv:2605.05876](https://arxiv.org/abs/2605.05876) (arXiv) — We present 3D Surface Splatting (3DSS), the first differentiable surface splatting renderer for physically-based inverse rendering from multi-view images. Our central

- **3DSS** [arXiv:2605.05876](https://arxiv.org/abs/2605.05876) (arXiv) — We present 3D Surface Splatting (3DSS), the first differentiable surface splatting renderer for physically-based inverse rendering from multi-view images. Our central

- **AmbiSuR** [arXiv:2605.12494](https://arxiv.org/abs/2605.12494) (ICML 2026) — Photometric ambiguity disambiguation for accurate GS surface reconstruction

- **AmbiSuR** [arXiv:2605.12494](https://arxiv.org/abs/2605.12494) (ICML 2026) — Photometric ambiguity disambiguation for accurate GS surface reconstruction

- **AnyGS2Mesh** [arXiv:2609.03304](https://arxiv.org/abs/2609.03304) (arXiv 2026) — Feed-forward mesh reconstruction from 3DGS with arbitrary-resolution views; Gaussian-guided transformer + TSDF fusion

- **AnyGS2Mesh** [arXiv:2609.03304](https://arxiv.org/abs/2609.03304) (arXiv 2026) — Feed-forward mesh reconstruction from 3DGS with arbitrary-resolution views; Gaussian-guided transformer + TSDF fusion

- **BEAST3D** [arXiv:2606.02937](https://arxiv.org/abs/2606.02937) (arXiv 2026) — We address these limitations with BEAST3D, a self-supervised pretraining framework that learns 3D visual representations from unlabeled, calibrated multi-view

- **BEAST3D** [arXiv:2606.02937](https://arxiv.org/abs/2606.02937) (arXiv 2026) — We address these limitations with BEAST3D, a self-supervised pretraining framework that learns 3D visual representations from unlabeled, calibrated multi-view

- **CF-Splat** [arXiv:2610.22108](https://arxiv.org/abs/2610.22108) (arXiv preprint) — Abstract context: Splatting-based radiance-field representations have recently gained popularity in robotic perception, enabling high-quality novel-view 

- **CF-Splat** [arXiv:2610.22108](https://arxiv.org/abs/2610.22108) (arXiv preprint) — Abstract context: Splatting-based radiance-field representations have recently gained popularity in robotic perception, enabling high-quality novel-view 

- **CLEAR-NeRF** [arXiv:2605.28125](https://arxiv.org/abs/2605.28125) (arXiv) — Collinearity and local-region enhanced accurate 3D reconstruction in unbounded scenes via geometric regularization

- **CLEAR-NeRF** [arXiv:2605.28125](https://arxiv.org/abs/2605.28125) (arXiv) — Collinearity and local-region enhanced accurate 3D reconstruction in unbounded scenes via geometric regularization

- **CoMVS-GS** [arXiv:2608.18413](https://arxiv.org/abs/2608.18413) (arXiv 2026) — General surface reconstruction framework that combines Multi-View Stereo with Gaussian splatting

- **CoMVS-GS** [arXiv:2608.18413](https://arxiv.org/abs/2608.18413) (arXiv 2026) — General surface reconstruction framework that combines Multi-View Stereo with Gaussian splatting

- **CubicSplat** [arXiv:2608.20803](https://arxiv.org/abs/2608.20803) (ECCV 2026) — Differentiable vector graphics via error-bounded forward relaxation; uniform polyline proxy replacing Bezier nearest-point solver; O(S^-2) bounded geometric

- **CubicSplat** [arXiv:2608.20803](https://arxiv.org/abs/2608.20803) (ECCV 2026) — Differentiable vector graphics via error-bounded forward relaxation; uniform polyline proxy replacing Bezier nearest-point solver; O(S^-2) bounded geometric

- **CVTSplat** [arXiv:2610.35347](https://arxiv.org/abs/2610.35347) (arXiv 2026) — Adaptive centroidal Voronoi tessellation for multi-view reconstruction; evolving-SDF-modulated cell sizes concentrate sites near the zero level set, with spe

- **CVTSplat** [arXiv:2610.35347](https://arxiv.org/abs/2610.35347) (arXiv 2026) — Adaptive centroidal Voronoi tessellation for multi-view reconstruction; evolving-SDF-modulated cell sizes concentrate sites near the zero level set, with spe

- **Deformable Triangle Splatting** [arXiv:2607.22446](https://arxiv.org/abs/2607.22446) (ECCV 2026) — Flexible primitives for real-time radiance field rendering; K control points per edge with learnable scalar displacement for non-convex s

- **Deformable Triangle Splatting** [arXiv:2607.22446](https://arxiv.org/abs/2607.22446) (ECCV 2026) — Flexible primitives for real-time radiance field rendering; K control points per edge with learnable scalar displacement for non-convex s

- **DinoComplete** [arXiv:2605.26949](https://arxiv.org/abs/2605.26949) (arXiv) — 3D shape completion with DINOv2-distilled semantic priors + state space model

- **DinoComplete** [arXiv:2605.26949](https://arxiv.org/abs/2605.26949) (arXiv) — 3D shape completion with DINOv2-distilled semantic priors + state space model

- **Direct Conditional Parameterization for N-Dimensional Splatting** [arXiv:2610.22778](https://arxiv.org/abs/2610.22778) (arXiv preprint) — Abstract context: N-dimensional splatting extends 3D Gaussian Splatting with view direction and ti

- **Direct Conditional Parameterization for N-Dimensional Splatting** [arXiv:2610.22778](https://arxiv.org/abs/2610.22778) (arXiv preprint) — Abstract context: N-dimensional splatting extends 3D Gaussian Splatting with view direction and ti

- **DySurface** [arXiv:2605.10360](https://arxiv.org/abs/2605.10360) (arXiv) — While novel view synthesis (NVS) for dynamic scenes has seen significant progress, reconstructing temporally consistent geometric surfaces remains a challenge. N

- **DySurface** [arXiv:2605.10360](https://arxiv.org/abs/2605.10360) (arXiv) — While novel view synthesis (NVS) for dynamic scenes has seen significant progress, reconstructing temporally consistent geometric surfaces remains a challenge. N

- **Elastic Triangle Splatting** [arXiv:2608.29106](https://arxiv.org/abs/2608.29106) (arXiv 2026) — While neural rendering methods such as 3D Gaussian Splatting achieve remarkable visual fidelity, traditional polygonal meshes remain the ba

- **Elastic Triangle Splatting** [arXiv:2608.29106](https://arxiv.org/abs/2608.29106) (arXiv 2026) — While neural rendering methods such as 3D Gaussian Splatting achieve remarkable visual fidelity, traditional polygonal meshes remain the ba

- **G2ARD-GS** [arXiv:2608.05704](https://arxiv.org/abs/2608.05704) (arXiv 2026) — Geometry-guided distillation method that converts a dense Gaussian prior instantiated either as a training-free point-cloud lift or a trained GS model into a

- **G2ARD-GS** [arXiv:2608.05704](https://arxiv.org/abs/2608.05704) (arXiv 2026) — Geometry-guided distillation method that converts a dense Gaussian prior instantiated either as a training-free point-cloud lift or a trained GS model into a

- **G2SR** [arXiv:2607.14470](https://arxiv.org/abs/2607.14470) (arXiv 2026) — G2SR, which exploits a well-posed core of the task: given cross-view 2D splat correspondences, 3D splats follow analytically from multi-view geometry

- **G2SR** [arXiv:2607.14470](https://arxiv.org/abs/2607.14470) (arXiv 2026) — G2SR, which exploits a well-posed core of the task: given cross-view 2D splat correspondences, 3D splats follow analytically from multi-view geometry

- **GaussFusion** [arXiv:2607.05906](https://arxiv.org/abs/2607.05906) (arXiv 2026) — Multimodal pre-training framework for 3D Gaussian representations

- **GaussFusion** [arXiv:2607.05906](https://arxiv.org/abs/2607.05906) (arXiv 2026) — Multimodal pre-training framework for 3D Gaussian representations

- **Gaussian-JEPA** [arXiv:2608.15651](https://arxiv.org/abs/2608.15651) (arXiv 2026) — Predicts representations of held-out Gaussian token blocks from visible context

- **Gaussian-JEPA** [arXiv:2608.15651](https://arxiv.org/abs/2608.15651) (arXiv 2026) — Predicts representations of held-out Gaussian token blocks from visible context

- **Gaussian Sculpting** [arXiv:2608.10602](https://arxiv.org/abs/2608.10602) (arXiv 2026) — To address these issues, we propose Gaussian Sculpting, a fully differentiable end-to-end framework for high-quality surface reconstruction

- **Gaussian Sculpting** [arXiv:2608.10602](https://arxiv.org/abs/2608.10602) (arXiv 2026) — To address these issues, we propose Gaussian Sculpting, a fully differentiable end-to-end framework for high-quality surface reconstruction

- **Geometry Gaussians** [arXiv:2606.05124](https://arxiv.org/abs/2606.05124) (arXiv) — Adds geometry opacity parameter per splat to decouple appearance and geometry in 3DGS

- **Geometry Gaussians** [arXiv:2606.05124](https://arxiv.org/abs/2606.05124) (arXiv) — Adds geometry opacity parameter per splat to decouple appearance and geometry in 3DGS

- **GLINT** [arXiv:2603.26181](https://arxiv.org/abs/2603.26181) (arXiv) — Scene-scale transparency via decomposed Gaussian radiance transport for glass

- **GLINT** [arXiv:2603.26181](https://arxiv.org/abs/2603.26181) (arXiv) — Scene-scale transparency via decomposed Gaussian radiance transport for glass

- **GOF** [arXiv:2312.13299](https://arxiv.org/abs/2312.13299) (ECCV 2024) — Gaussian Opacity Field: opacity-weighted TSDF fusion for high-fidelity surface extraction from GS

- **GOF** [arXiv:2312.13299](https://arxiv.org/abs/2312.13299) (ECCV 2024) — Gaussian Opacity Field: opacity-weighted TSDF fusion for high-fidelity surface extraction from GS

- **GOFish** [arXiv:2610.37818](https://arxiv.org/abs/2610.37818) (arXiv 2026) — First 3D Gaussian Splatting surface reconstruction method to train directly on fisheye images, rasterize FoVs beyond 180°, and extract an aligned mesh from the

- **GOFish** [arXiv:2610.37818](https://arxiv.org/abs/2610.37818) (arXiv 2026) — First 3D Gaussian Splatting surface reconstruction method to train directly on fisheye images, rasterize FoVs beyond 180°, and extract an aligned mesh from the

- **GS2Mesh** [arXiv:2404.01810](https://arxiv.org/abs/2404.01810) (CVPR 2024) — Surface-regularized GS → mesh extraction with multi-view depth consistency constraints

- **GS2Mesh** [arXiv:2404.01810](https://arxiv.org/abs/2404.01810) (CVPR 2024) — Surface-regularized GS → mesh extraction with multi-view depth consistency constraints

- **GS-CPE** [arXiv:2608.10938](https://arxiv.org/abs/2608.10938) (arXiv 2026) — This study introduces GS-CPE (Gaussian Splatting based Camera Pose Estimation), a coarse-to-fine framework for 6-DoF camera pose estimation that unifies geomet

- **GS-CPE** [arXiv:2608.10938](https://arxiv.org/abs/2608.10938) (arXiv 2026) — This study introduces GS-CPE (Gaussian Splatting based Camera Pose Estimation), a coarse-to-fine framework for 6-DoF camera pose estimation that unifies geomet

- **GSDF** [arXiv:2403.16964](https://arxiv.org/abs/2403.16964) (NeurIPS 2024) — Dual representation: GS guides SDF geometry, SDF provides normal regularization for GS

- **GSDF** [arXiv:2403.16964](https://arxiv.org/abs/2403.16964) (NeurIPS 2024) — Dual representation: GS guides SDF geometry, SDF provides normal regularization for GS

- **GS-Manifold** [no arXiv ID; verify primary source] (CVPR 2025) — Manifold-constrained Gaussians for surface reconstruction with topological guarantees - **GSurf** [arXiv:2411.15723](https://arxiv.org/abs/2411.15723) (CVPR 2024) — Gaussi

- **GS-Manifold** [no arXiv ID; verify primary source] (CVPR 2025) — Manifold-constrained Gaussians for surface reconstruction with topological guarantees - **GSurf** [arXiv:2411.15723](https://arxiv.org/abs/2411.15723) (CVPR 2024) — Gaussi

- **GSurf** [arXiv:2411.15723](https://arxiv.org/abs/2411.15723) (CVPR 2024) — Gaussian surface reconstruction with SDF-GS hybrid representation for watertight meshes

- **GSurf** [arXiv:2411.15723](https://arxiv.org/abs/2411.15723) (CVPR 2024) — Gaussian surface reconstruction with SDF-GS hybrid representation for watertight meshes

- **GVKF** [arXiv:2411.01853](https://arxiv.org/abs/2411.01853) (NeurIPS 2024) — Gaussian Voxel Kernel Functions for highly efficient surface reconstruction via TSDF fusion

- **GVKF** [arXiv:2411.01853](https://arxiv.org/abs/2411.01853) (NeurIPS 2024) — Gaussian Voxel Kernel Functions for highly efficient surface reconstruction via TSDF fusion

- **Heat Kernel Textures** [no arXiv ID; verify primary source] (ECCV 2026 Best Paper) — Geodesic Gaussians that do not splat; anisotropic heat kernel on non-Euclidean manifolds replacing standard Euclidean Gaussians for surface texture rep

- **Heat Kernel Textures** [no arXiv ID; verify primary source] (ECCV 2026 Best Paper) — Geodesic Gaussians that do not splat; anisotropic heat kernel on non-Euclidean manifolds replacing standard Euclidean Gaussians for surface texture rep

- **HiCo-GS** [arXiv:2608.14136](https://arxiv.org/abs/2608.14136) (arXiv 2026) — High-fidelity reconstruction framework with two complementary modules

- **HiCo-GS** [arXiv:2608.14136](https://arxiv.org/abs/2608.14136) (arXiv 2026) — High-fidelity reconstruction framework with two complementary modules

- **HoloTetSphere** [arXiv:2607.08398](https://arxiv.org/abs/2607.08398) (ECCV 2026) — Unified TetSphere mesh reconstruction for physical simulation; Gaussian sphere coupling + edge-based element pruning; bypasses error-prone tetrahedraliza

- **HoloTetSphere** [arXiv:2607.08398](https://arxiv.org/abs/2607.08398) (ECCV 2026) — Unified TetSphere mesh reconstruction for physical simulation; Gaussian sphere coupling + edge-based element pruning; bypasses error-prone tetrahedraliza

- **Incremental 3D Gaussian Triangulation** [no arXiv ID; verify primary source] (arXiv) — Online incremental mesh extraction from dense Gaussian triangulation; plane-based pulling constraint + dynamic historical region freezing - **LeanGau

- **Incremental 3D Gaussian Triangulation** [no arXiv ID; verify primary source] (arXiv) — Online incremental mesh extraction from dense Gaussian triangulation; plane-based pulling constraint + dynamic historical region freezing - **LeanGau

- **LeanGaussian** [no arXiv ID; verify primary source] (CVPR 2025) — Extreme compression from single RGB image for efficient large-scale rendering - **LiteTex-GS** [arXiv:2609.23380](https://arxiv.org/abs/2609.23380) (arXiv 2026) — Lightwe

- **LeanGaussian** [no arXiv ID; verify primary source] (CVPR 2025) — Extreme compression from single RGB image for efficient large-scale rendering - **LiteTex-GS** [arXiv:2609.23380](https://arxiv.org/abs/2609.23380) (arXiv 2026) — Lightwe

- **LiteTex-GS** [arXiv:2609.23380](https://arxiv.org/abs/2609.23380) (arXiv 2026) — Lightweight texturing for Gaussian Splatting that attaches texture maps to primitives to reduce memory and optimization cost of high-frequency details.

- **LiteTex-GS** [arXiv:2609.23380](https://arxiv.org/abs/2609.23380) (arXiv 2026) — Lightweight texturing for Gaussian Splatting that attaches texture maps to primitives to reduce memory and optimization cost of high-frequency details.

- **Manifold-GS** [arXiv:2608.00214](https://arxiv.org/abs/2608.00214) (arXiv 2026) — Certified hybrid asset layer for Gaussian scenes

- **Manifold-GS** [arXiv:2608.00214](https://arxiv.org/abs/2608.00214) (arXiv 2026) — Certified hybrid asset layer for Gaussian scenes

- **Mesh2GS** [arXiv:2606.21898](https://arxiv.org/abs/2606.21898) (arXiv 2026) — Instead of relying on heuristic strategies that bind 3D Gaussians to the mesh, we propose a novel white-box 3DGS construction framework, termed Mesh2GS

- **Mesh2GS** [arXiv:2606.21898](https://arxiv.org/abs/2606.21898) (arXiv 2026) — Instead of relying on heuristic strategies that bind 3D Gaussians to the mesh, we propose a novel white-box 3DGS construction framework, termed Mesh2GS

- **MeshWeaver** [arXiv:2606.04688](https://arxiv.org/abs/2606.04688) (CVPR) — Autoregressive next-vertex prediction guided by multi-level sparse-voxel encoder; 18% compression ratio

- **MeshWeaver** [arXiv:2606.04688](https://arxiv.org/abs/2606.04688) (CVPR) — Autoregressive next-vertex prediction guided by multi-level sparse-voxel encoder; 18% compression ratio

- **NegGS** [arXiv:2405.18163](https://arxiv.org/abs/2405.18163) (arXiv) — Negative color values for ring/crescent/non-convex structures

- **NegGS** [arXiv:2405.18163](https://arxiv.org/abs/2405.18163) (arXiv) — Negative color values for ring/crescent/non-convex structures

- **Neural Gabor Splatting** [arXiv:2604.15941](https://arxiv.org/abs/2604.15941) (CVPR 2026) — Recent years have witnessed the rapid emergence of 3D Gaussian splatting (3DGS) as a powerful approach for 3D reconstruction and novel view synt

- **Neural Gabor Splatting** [arXiv:2604.15941](https://arxiv.org/abs/2604.15941) (CVPR 2026) — Recent years have witnessed the rapid emergence of 3D Gaussian splatting (3DGS) as a powerful approach for 3D reconstruction and novel view synt

- **Neural Harmonic Textures** [arXiv:2604.01204](https://arxiv.org/abs/2604.01204) (ECCV 2026) — High-quality primitive-based neural reconstruction; per-primitive learnable feature scaffolds with periodic activation; harmonic decomposition

- **Neural Harmonic Textures** [arXiv:2604.01204](https://arxiv.org/abs/2604.01204) (ECCV 2026) — High-quality primitive-based neural reconstruction; per-primitive learnable feature scaffolds with periodic activation; harmonic decomposition

- **NeuSG** [arXiv:2312.00846](https://arxiv.org/abs/2312.00846) (CVPR 2024) — Neural surface-guided GS: SDF-guided Gaussian anchoring for consistent surface reconstruction

- **NeuSG** [arXiv:2312.00846](https://arxiv.org/abs/2312.00846) (CVPR 2024) — Neural surface-guided GS: SDF-guided Gaussian anchoring for consistent surface reconstruction

- **ObjectSplat** [arXiv:2608.30423](https://arxiv.org/abs/2608.30423) (arXiv) — Object-level mesh splatting for enhanced mesh fidelity and interactivity; bridges 3DGS and mesh-based representations at object granularity

- **ObjectSplat** [arXiv:2608.30423](https://arxiv.org/abs/2608.30423) (arXiv) — Object-level mesh splatting for enhanced mesh fidelity and interactivity; bridges 3DGS and mesh-based representations at object granularity

- **P2M++** [arXiv:2605.00429](https://arxiv.org/abs/2605.00429) (arXiv) — Enhanced solver for point-to-mesh distance queries

- **P2M++** [arXiv:2605.00429](https://arxiv.org/abs/2605.00429) (arXiv) — Enhanced solver for point-to-mesh distance queries

- **PAGaS** [arXiv:2604.22129](https://arxiv.org/abs/2604.22129) (arXiv) — Pixel-aligned 1DoF Gaussians for depth refinement

- **PAGaS** [arXiv:2604.22129](https://arxiv.org/abs/2604.22129) (arXiv) — Pixel-aligned 1DoF Gaussians for depth refinement

- **PCAsplat** [arXiv:2610.11011](https://arxiv.org/abs/2610.11011) (arXiv 2026) — Geometry-aware regularization framework for Gaussian splatting based on differentiable local PCA that updates occluded Gaussians, regularizes eigenvalues tow

- **PCAsplat** [arXiv:2610.11011](https://arxiv.org/abs/2610.11011) (arXiv 2026) — Geometry-aware regularization framework for Gaussian splatting based on differentiable local PCA that updates occluded Gaussians, regularizes eigenvalues tow

- **PDE-Constrained 3DGS** [no arXiv ID; verify primary source] (CVPR) — PDE physical constraints for 3DGS artifact removal; geometry occupancy supervision for 3D consistency - **PGSR** [arXiv:2406.06521](https://arxiv.org/abs/2406.06521) (

- **PDE-Constrained 3DGS** [no arXiv ID; verify primary source] (CVPR) — PDE physical constraints for 3DGS artifact removal; geometry occupancy supervision for 3D consistency - **PGSR** [arXiv:2406.06521](https://arxiv.org/abs/2406.06521) (

- **PGSR** [arXiv:2406.06521](https://arxiv.org/abs/2406.06521) (TVCG 2024) — Planar-based regularizer for high-fidelity surface reconstruction

- **PGSR** [arXiv:2406.06521](https://arxiv.org/abs/2406.06521) (TVCG 2024) — Planar-based regularizer for high-fidelity surface reconstruction

- **PointSplat** [arXiv:2604.09903](https://arxiv.org/abs/2604.09903) (arXiv) — Geometry-driven pruning + Transformer refinement for efficient 3DGS

- **PointSplat** [arXiv:2604.09903](https://arxiv.org/abs/2604.09903) (arXiv) — Geometry-driven pruning + Transformer refinement for efficient 3DGS

- **QuerySplat** [arXiv:2608.01186](https://arxiv.org/abs/2608.01186) (arXiv 2026) — To overcome these deficiencies, we propose textbfQuerySplat, a feed-forward 3DGS framework driven by geometric priors and explicit appearance decoupling

- **QuerySplat** [arXiv:2608.01186](https://arxiv.org/abs/2608.01186) (arXiv 2026) — To overcome these deficiencies, we propose textbfQuerySplat, a feed-forward 3DGS framework driven by geometric priors and explicit appearance decoupling

- **RGS (Reflection)** [arXiv:2609.19421](https://arxiv.org/abs/2609.19421) (arXiv 2026) — Physically-based deferred rendering for reflection-aware Gaussian Splatting, modeling specular regions on reflective objects.

- **RGS (Reflection)** [arXiv:2609.19421](https://arxiv.org/abs/2609.19421) (arXiv 2026) — Physically-based deferred rendering for reflection-aware Gaussian Splatting, modeling specular regions on reflective objects.

- **RT-Splatting** [arXiv:2605.18263](https://arxiv.org/abs/2605.18263) (CVPR 2026 Highlight) — Joint reflection-transmission modeling; geometric occupancy disentangled from optical opacity per Gaussian

- **RT-Splatting** [arXiv:2605.18263](https://arxiv.org/abs/2605.18263) (CVPR 2026 Highlight) — Joint reflection-transmission modeling; geometric occupancy disentangled from optical opacity per Gaussian

- **SAGS** [arXiv:2510.27318](https://arxiv.org/abs/2510.27318) (ECCV 2024) — Shape-aware GS: shape priors guiding Gaussian distribution for anatomically faithful reconstruction

- **SAGS** [arXiv:2510.27318](https://arxiv.org/abs/2510.27318) (ECCV 2024) — Shape-aware GS: shape priors guiding Gaussian distribution for anatomically faithful reconstruction

- **SAND** [no arXiv ID; verify primary source] (SIGGRAPH 2026 TOG) — Spatially adaptive network depth for efficient neural implicit surface sampling - **SatSurfGS** [arXiv:2605.07181](https://arxiv.org/abs/2605.07181) (arXiv 2026) — Genera

- **SAND** [no arXiv ID; verify primary source] (SIGGRAPH 2026 TOG) — Spatially adaptive network depth for efficient neural implicit surface sampling - **SatSurfGS** [arXiv:2605.07181](https://arxiv.org/abs/2605.07181) (arXiv 2026) — Genera

- **SatSurfGS** [arXiv:2605.07181](https://arxiv.org/abs/2605.07181) (arXiv 2026) — Generalizable 2DGS for sparse-view satellite surface reconstruction; confidence-aware multi-view

- **SatSurfGS** [arXiv:2605.07181](https://arxiv.org/abs/2605.07181) (arXiv 2026) — Generalizable 2DGS for sparse-view satellite surface reconstruction; confidence-aware multi-view

- **SparseOIT** [arXiv:2605.13855](https://arxiv.org/abs/2605.13855) (arXiv 2026) — Order-independent transparency via active set method for glass/refractive scenes

- **SparseOIT** [arXiv:2605.13855](https://arxiv.org/abs/2605.13855) (arXiv 2026) — Order-independent transparency via active set method for glass/refractive scenes

- **SplashSplat** [arXiv:2609.20818](https://arxiv.org/abs/2609.20818) (arXiv 2026) — Built on a single principle: impose physical structure only where the observations can constrain it

- **SplashSplat** [arXiv:2609.20818](https://arxiv.org/abs/2609.20818) (arXiv 2026) — Built on a single principle: impose physical structure only where the observations can constrain it

- **StructureGS** [arXiv:2607.26889](https://arxiv.org/abs/2607.26889) (arXiv 2026) — To address this limitation, we introduce StructureGS, a reconstruction framework for articulated objects that integrates structure-aware guidance into 3D 

- **StructureGS** [arXiv:2607.26889](https://arxiv.org/abs/2607.26889) (arXiv 2026) — To address this limitation, we introduce StructureGS, a reconstruction framework for articulated objects that integrates structure-aware guidance into 3D 

- **SuGaR** [arXiv:2311.12775](https://arxiv.org/abs/2311.12775) (CVPR 2024) — Surface-aligned Gaussians for mesh extraction via TSDF + Marching Cubes

- **SuGaR** [arXiv:2311.12775](https://arxiv.org/abs/2311.12775) (CVPR 2024) — Surface-aligned Gaussians for mesh extraction via TSDF + Marching Cubes

- **SuperGS** [arXiv:2410.02571](https://arxiv.org/abs/2410.02571) (CVPR 2024) — Super-resolution guided GS: using 2D SR priors to enhance 3DGS rendering quality

- **SuperGS** [arXiv:2410.02571](https://arxiv.org/abs/2410.02571) (CVPR 2024) — Super-resolution guided GS: using 2D SR priors to enhance 3DGS rendering quality

- **SVGS** [arXiv:2411.18966](https://arxiv.org/abs/2411.18966) (arXiv) — Gaussian Splatting demonstrates impressive results in multi-view reconstruction based on Gaussian explicit representations. However, the current Gaussian primitives o

- **SVGS** [arXiv:2411.18966](https://arxiv.org/abs/2411.18966) (arXiv) — Gaussian Splatting demonstrates impressive results in multi-view reconstruction based on Gaussian explicit representations. However, the current Gaussian primitives o

- **TetraSDF** [arXiv:2511.16273](https://arxiv.org/abs/2511.16273) (ECCV 2026) — Analytic isosurface extraction with multi-resolution tetrahedral grid; ReLU MLP composed with tetrahedral position encoding preserves CPWA structure for exact

- **TetraSDF** [arXiv:2511.16273](https://arxiv.org/abs/2511.16273) (ECCV 2026) — Analytic isosurface extraction with multi-resolution tetrahedral grid; ReLU MLP composed with tetrahedral position encoding preserves CPWA structure for exact

- **TopoSurfel** [arXiv:2608.20687](https://arxiv.org/abs/2608.20687) (arXiv 2026) — To address this limitation, we propose TopoSurfel, a novel framework that closes the loop between Gaussian surfels and continuous meshes

- **TopoSurfel** [arXiv:2608.20687](https://arxiv.org/abs/2608.20687) (arXiv 2026) — To address this limitation, we propose TopoSurfel, a novel framework that closes the loop between Gaussian surfels and continuous meshes

- **TransmissiveGS** [arXiv:2605.10705](https://arxiv.org/abs/2605.10705) (arXiv) — Residual-guided disentangled Gaussian Splatting for transmissive scene reconstruction; separates reflection and transmission components via residual-guided 

- **TransmissiveGS** [arXiv:2605.10705](https://arxiv.org/abs/2605.10705) (arXiv) — Residual-guided disentangled Gaussian Splatting for transmissive scene reconstruction; separates reflection and transmission components via residual-guided 

- **TRiGS** [arXiv:2604.00538](https://arxiv.org/abs/2604.00538) (CVPR 2024) — Tri-plane augmented Gaussian Splatting: tri-plane features + Gaussian geometry for hybrid representation

- **TRiGS** [arXiv:2604.00538](https://arxiv.org/abs/2604.00538) (CVPR 2024) — Tri-plane augmented Gaussian Splatting: tri-plane features + Gaussian geometry for hybrid representation

- **VCR-GauS** [arXiv:2406.05774](https://arxiv.org/abs/2406.05774) (NeurIPS 2024) — View-consistent depth-normal regularization for GS surface reconstruction

- **VCR-GauS** [arXiv:2406.05774](https://arxiv.org/abs/2406.05774) (NeurIPS 2024) — View-consistent depth-normal regularization for GS surface reconstruction

- **View-Dependent Splatting** [arXiv:2605.25426](https://arxiv.org/abs/2605.25426) (SIGGRAPH) — We present a differentiable framework to automatically learn view-dependent 2D kernels in a splatting-based pipeline to improve reconstruction 

- **View-Dependent Splatting** [arXiv:2605.25426](https://arxiv.org/abs/2605.25426) (SIGGRAPH) — We present a differentiable framework to automatically learn view-dependent 2D kernels in a splatting-based pipeline to improve reconstruction 

- **VoxelGS** [arXiv:2605.26616](https://arxiv.org/abs/2605.26616) (arXiv) — Scaffold-anchored Gaussians tethered to voxelized SDF for fast monocular surface reconstruction

- **VoxelGS** [arXiv:2605.26616](https://arxiv.org/abs/2605.26616) (arXiv) — Scaffold-anchored Gaussians tethered to voxelized SDF for fast monocular surface reconstruction

- **2DGS** [arXiv:2403.17888](https://arxiv.org/abs/2403.17888) (SIGGRAPH 2024) — 2D disks on surfaces enabling direct mesh extraction via Poisson reconstruction

- **2DGS** [arXiv:2403.17888](https://arxiv.org/abs/2403.17888) (SIGGRAPH 2024) — 2D disks on surfaces enabling direct mesh extraction via Poisson reconstruction

- **3DCodeBench** [arXiv:2606.01057](https://arxiv.org/abs/2606.01057) (arXiv 2026) — Systematic benchmark for VLM agents on procedural 3D generation (text/image -> code -> 3D); evaluates 12 VLMs; includes 3DCodeArena human preference ranki

- **3DCodeBench** [arXiv:2606.01057](https://arxiv.org/abs/2606.01057) (arXiv 2026) — Systematic benchmark for VLM agents on procedural 3D generation (text/image -> code -> 3D); evaluates 12 VLMs; includes 3DCodeArena human preference ranki

- **ASSEMCAD** [no arXiv ID; verify primary source] (arXiv) — Axiom-driven framework for production-ready CAD assembly generation from natural language; port-and-mate-based library with executable mate relations + engineering axioms; outper

- **ASSEMCAD** [no arXiv ID; verify primary source] (arXiv) — Axiom-driven framework for production-ready CAD assembly generation from natural language; port-and-mate-based library with executable mate relations + engineering axioms; outper

- **BRepCLIP** [arXiv:2606.05515](https://arxiv.org/abs/2606.05515) (arXiv 2026) — Contrastive multi-modal pre-training with BRep primitives for CAD understanding; cross-modal retrieval

- **BRepCLIP** [arXiv:2606.05515](https://arxiv.org/abs/2606.05515) (arXiv 2026) — Contrastive multi-modal pre-training with BRep primitives for CAD understanding; cross-modal retrieval

- **BrepGaussian** [arXiv:2602.21105](https://arxiv.org/abs/2602.21105) (CVPR 2026) — 3DGS + B-rep CAD reconstruction to parametric STEP models

- **BrepGaussian** [arXiv:2602.21105](https://arxiv.org/abs/2602.21105) (CVPR 2026) — 3DGS + B-rep CAD reconstruction to parametric STEP models

- **BrepGiff** [no arXiv ID; verify primary source] (arXiv 2026) — Lightweight generation of complex B-rep with 3D GAT Diffusion; graph attention network diffusion for B-rep topology+geometry (IEEE Xplore) - **CADDreamer** [no arXiv ID; ver

- **BrepGiff** [no arXiv ID; verify primary source] (arXiv 2026) — Lightweight generation of complex B-rep with 3D GAT Diffusion; graph attention network diffusion for B-rep topology+geometry (IEEE Xplore) - **CADDreamer** [no arXiv ID; ver

- **CADDreamer** [no arXiv ID; verify primary source] (CVPR 2025 Highlight) — CAD object generation from single-view images; parametric primitive sequence generation from 2D input - **CADFit** [arXiv:2605.01171](https://arxiv.org/abs/2605.0

- **CADDreamer** [no arXiv ID; verify primary source] (CVPR 2025 Highlight) — CAD object generation from single-view images; parametric primitive sequence generation from 2D input - **CADFit** [arXiv:2605.01171](https://arxiv.org/abs/2605.0

- **CADFit** [arXiv:2605.01171](https://arxiv.org/abs/2605.01171) (arXiv) — Despite recent progress, recovering parametric CAD construction sequences from geometric input, such as meshes or point clouds, is a key challenge for design and ma

- **CADFit** [arXiv:2605.01171](https://arxiv.org/abs/2605.01171) (arXiv) — Despite recent progress, recovering parametric CAD construction sequences from geometric input, such as meshes or point clouds, is a key challenge for design and ma

- **CADFS** [arXiv:2605.01925](https://arxiv.org/abs/2605.01925) (CVPR 2026) — Large-scale CAD program dataset + LLM-assisted CAD understanding

- **CADFS** [arXiv:2605.01925](https://arxiv.org/abs/2605.01925) (CVPR 2026) — Large-scale CAD program dataset + LLM-assisted CAD understanding

- **CADSplat** [arXiv:2609.18473](https://arxiv.org/abs/2609.18473) (arXiv 2026) — Framework that reconstructs photorealistic, geometrically accurate digital twins from sparse (<15 views), wide-baseline posed images of an object by regulari

- **CADSplat** [arXiv:2609.18473](https://arxiv.org/abs/2609.18473) (arXiv 2026) — Framework that reconstructs photorealistic, geometrically accurate digital twins from sparse (<15 views), wide-baseline posed images of an object by regulari

- **KDH-CAD** [arXiv:2606.01702](https://arxiv.org/abs/2606.01702) (arXiv 2026) — Knowledge-data hybrid CAD learning: pretrained foundation model + structured domain knowledge from textbooks + small labeled CAD data; 92.6% accuracy with onl

- **KDH-CAD** [arXiv:2606.01702](https://arxiv.org/abs/2606.01702) (arXiv 2026) — Knowledge-data hybrid CAD learning: pretrained foundation model + structured domain knowledge from textbooks + small labeled CAD data; 92.6% accuracy with onl

- **MaGS** [no arXiv ID; verify primary source] (ICCV 2025) — Mesh-adsorbed Gaussians; deform mesh → Gaussians follow - **MidSurfNet** [arXiv:2606.01891](https://arxiv.org/abs/2606.01891) (arXiv 2026) — Learning-augmented mid-surface abstra

- **MaGS** [no arXiv ID; verify primary source] (ICCV 2025) — Mesh-adsorbed Gaussians; deform mesh → Gaussians follow - **MidSurfNet** [arXiv:2606.01891](https://arxiv.org/abs/2606.01891) (arXiv 2026) — Learning-augmented mid-surface abstra

- **MidSurfNet** [arXiv:2606.01891](https://arxiv.org/abs/2606.01891) (arXiv 2026) — Learning-augmented mid-surface abstraction: neural face pairing module + interference implicit field (SDF intersection) for arbitrary offset control; 87.32

- **MidSurfNet** [arXiv:2606.01891](https://arxiv.org/abs/2606.01891) (arXiv 2026) — Learning-augmented mid-surface abstraction: neural face pairing module + interference implicit field (SDF intersection) for arbitrary offset control; 87.32

- **RealCAD** [arXiv:2608.30617](https://arxiv.org/abs/2608.30617) (arXiv) — Real-world image to CAD reconstruction; handles domain gap and parametric deviation from imperfect photographs; code and dataset released

- **RealCAD** [arXiv:2608.30617](https://arxiv.org/abs/2608.30617) (arXiv) — Real-world image to CAD reconstruction; handles domain gap and parametric deviation from imperfect photographs; code and dataset released

- **RORA** [arXiv:2608.04842](https://arxiv.org/abs/2608.04842) (arXiv) — Single static video to articulated object end-to-end pipeline; mixed 3DGS plus mesh; convex decomposition plus automatic joint suggestion algorithm; deployed in Unrea

- **RORA** [arXiv:2608.04842](https://arxiv.org/abs/2608.04842) (arXiv) — Single static video to articulated object end-to-end pipeline; mixed 3DGS plus mesh; convex decomposition plus automatic joint suggestion algorithm; deployed in Unrea

- **SEIG** [arXiv:2606.02580](https://arxiv.org/abs/2606.02580) (arXiv 2026) — VLM generates executable Blender programs from single image via staged decomposition (geometry->materials->composition->lighting); no differentiable rendering or

- **SEIG** [arXiv:2606.02580](https://arxiv.org/abs/2606.02580) (arXiv 2026) — VLM generates executable Blender programs from single image via staged decomposition (geometry->materials->composition->lighting); no differentiable rendering or

- **UniCAD** [arXiv:2606.05058](https://arxiv.org/abs/2606.05058) (arXiv) — Unified multi-modal multi-task CAD benchmark + UniCAD-MLLM for point/text/image to CAD reconstruction

- **UniCAD** [arXiv:2606.05058](https://arxiv.org/abs/2606.05058) (arXiv) — Unified multi-modal multi-task CAD benchmark + UniCAD-MLLM for point/text/image to CAD reconstruction

- **UniMGS** [no arXiv ID; verify primary source] (AAAI 2026) — Single-pass rasterization for both mesh and Gaussians simultaneously - **Vol3DGS** [no arXiv ID; verify primary source] (CVPR 2025) — Physically accurate volume-consistent rend

- **UniMGS** [no arXiv ID; verify primary source] (AAAI 2026) — Single-pass rasterization for both mesh and Gaussians simultaneously - **Vol3DGS** [no arXiv ID; verify primary source] (CVPR 2025) — Physically accurate volume-consistent rend

- **Zero-to-CAD** [arXiv:2604.24479](https://arxiv.org/abs/2604.24479) (arXiv 2026) — Agentic synthesis of parametric CAD programs; 2B VLM trained on 1M+ synthetic CAD programs

- **Zero-to-CAD** [arXiv:2604.24479](https://arxiv.org/abs/2604.24479) (arXiv 2026) — Agentic synthesis of parametric CAD programs; 2B VLM trained on 1M+ synthetic CAD programs

- **2DGS-Planner** [arXiv:2610.11752](https://arxiv.org/abs/2610.11752) (arXiv 2026) — Rasterization-based path planner for ground robots that reads planning-relevant geometry from a 2DGS map; multi-view attribution converts rendered normal

- **2DGS-Planner** [arXiv:2610.11752](https://arxiv.org/abs/2610.11752) (arXiv 2026) — Rasterization-based path planner for ground robots that reads planning-relevant geometry from a 2DGS map; multi-view attribution converts rendered normal

- **AirSplan** [arXiv:2609.21226](https://arxiv.org/abs/2609.21226) (arXiv 2026) — Adopts a normalized variant of 3D Gaussian Splatting that encodes high-fidelity scene geometry

- **AirSplan** [arXiv:2609.21226](https://arxiv.org/abs/2609.21226) (arXiv 2026) — Adopts a normalized variant of 3D Gaussian Splatting that encodes high-fidelity scene geometry

- **ArtGS** [no arXiv ID; verify primary source] (IEEE 2026) — 3D Gaussian Splatting for interactive visual-physical modeling and manipulation of articulated objects; bone-driven deformation + contact-aware rendering - **FastBridge** [arXiv

- **ArtGS** [no arXiv ID; verify primary source] (IEEE 2026) — 3D Gaussian Splatting for interactive visual-physical modeling and manipulation of articulated objects; bone-driven deformation + contact-aware rendering - **FastBridge** [arXiv

- **FastBridge** [arXiv:2607.01200](https://arxiv.org/abs/2607.01200) (arXiv 2026) — Fast quadrotor flight requires safe obstacle avoidance under tight onboard compute limits

- **FastBridge** [arXiv:2607.01200](https://arxiv.org/abs/2607.01200) (arXiv 2026) — Fast quadrotor flight requires safe obstacle avoidance under tight onboard compute limits

- **FlyMirage** [arXiv:2605.19600](https://arxiv.org/abs/2605.19600) (arXiv 2026) — To address these challenges, we introduce FlyMirage, a highly scalable and fully automated data generation pipeline for aerial VLN

- **FlyMirage** [arXiv:2605.19600](https://arxiv.org/abs/2605.19600) (arXiv 2026) — To address these challenges, we introduce FlyMirage, a highly scalable and fully automated data generation pipeline for aerial VLN

- **Forecast-GS** [arXiv:2605.11144](https://arxiv.org/abs/2605.11144) (arXiv) — We introduce Forecast-aware Gaussian Splatting (Forecast-GS), a predictive 3D representation framework for language-conditioned robotic manipulation. While rec

- **Forecast-GS** [arXiv:2605.11144](https://arxiv.org/abs/2605.11144) (arXiv) — We introduce Forecast-aware Gaussian Splatting (Forecast-GS), a predictive 3D representation framework for language-conditioned robotic manipulation. While rec

- **FreeArtGS** [arXiv:2603.22102](https://arxiv.org/abs/2603.22102) (arXiv 2026) — Articulated Gaussian Splatting under free-moving scenario; scalable articulated object reconstruction

- **FreeArtGS** [arXiv:2603.22102](https://arxiv.org/abs/2603.22102) (arXiv 2026) — Articulated Gaussian Splatting under free-moving scenario; scalable articulated object reconstruction

- **FreeMoCap** [no arXiv ID; verify primary source] (arXiv) — Open-source markerless motion capture from webcams (AGPL-3.0, 8.3k stars); drives 3DGS avatars; dual-use multi-cam for mocap + 3DGS training - **GaussianGrasper** [arXiv:2403.09

- **FreeMoCap** [no arXiv ID; verify primary source] (arXiv) — Open-source markerless motion capture from webcams (AGPL-3.0, 8.3k stars); drives 3DGS avatars; dual-use multi-cam for mocap + 3DGS training - **GaussianGrasper** [arXiv:2403.09

- **GaussianGrasper** [arXiv:2403.09637](https://arxiv.org/abs/2403.09637) (IEEE T-RO 2024) — Constructing a 3D scene capable of accommodating open-ended language queries, is a pivotal pursuit, particularly within the domain of robotics. Su

- **GaussianGrasper** [arXiv:2403.09637](https://arxiv.org/abs/2403.09637) (IEEE T-RO 2024) — Constructing a 3D scene capable of accommodating open-ended language queries, is a pivotal pursuit, particularly within the domain of robotics. Su

- **GaussLite** [arXiv:2606.30809](https://arxiv.org/abs/2606.30809) (arXiv 2026) — Task-driven 3DGS mapping system that conditions its representation density on a natural-language task specification

- **GaussLite** [arXiv:2606.30809](https://arxiv.org/abs/2606.30809) (arXiv 2026) — Task-driven 3DGS mapping system that conditions its representation density on a natural-language task specification

- **GaussMemory** [arXiv:2608.14986](https://arxiv.org/abs/2608.14986) (arXiv 2026) — Paradigm shift from passive storage to active, task-driven spatial memory

- **GaussMemory** [arXiv:2608.14986](https://arxiv.org/abs/2608.14986) (arXiv 2026) — Paradigm shift from passive storage to active, task-driven spatial memory

- **GaussNav** [arXiv:2403.11625](https://arxiv.org/abs/2403.11625) (CVPR 2024) — GS-based navigation with language-guided semantic Gaussian maps for embodied agents

- **GaussNav** [arXiv:2403.11625](https://arxiv.org/abs/2403.11625) (CVPR 2024) — GS-based navigation with language-guided semantic Gaussian maps for embodied agents

- **GEMS** [arXiv:2610.33158](https://arxiv.org/abs/2610.33158) (arXiv 2026) — GEMS, a framework that combines a persistent spatio-temporal 3D Gaussian Splatting (3DGS) map with an agentic reasoning pipeline

- **GEMS** [arXiv:2610.33158](https://arxiv.org/abs/2610.33158) (arXiv 2026) — GEMS, a framework that combines a persistent spatio-temporal 3D Gaussian Splatting (3DGS) map with an agentic reasoning pipeline

- **Geometry-Aware Monocular Flight** [arXiv:2610.26508](https://arxiv.org/abs/2610.26508) (arXiv preprint) — Monocular planning representations learned using privileged depth supervision and a 3DGS simulator; a robotics application rather 

- **Geometry-Aware Monocular Flight** [arXiv:2610.26508](https://arxiv.org/abs/2610.26508) (arXiv preprint) — Monocular planning representations learned using privileged depth supervision and a 3DGS simulator; a robotics application rather 

- **GIC** [arXiv:2406.14927](https://arxiv.org/abs/2406.14927) (NeurIPS 2024) — Gaussian-Informed Continuum for physical property identification and differentiable simulation

- **GIC** [arXiv:2406.14927](https://arxiv.org/abs/2406.14927) (NeurIPS 2024) — Gaussian-Informed Continuum for physical property identification and differentiable simulation

- **GraspSplats** [arXiv:2409.02084](https://arxiv.org/abs/2409.02084) (CoRL 2024) — The ability for robots to perform efficient and zero-shot grasping of object parts is crucial for practical applications and is becoming prevalent with rec

- **GraspSplats** [arXiv:2409.02084](https://arxiv.org/abs/2409.02084) (CoRL 2024) — The ability for robots to perform efficient and zero-shot grasping of object parts is crucial for practical applications and is becoming prevalent with rec

- **GSDrive** [arXiv:2604.28111](https://arxiv.org/abs/2604.28111) (arXiv) — Reinforcing driving policies with 3DGS environment for high-fidelity RL training

- **GSDrive** [arXiv:2604.28111](https://arxiv.org/abs/2604.28111) (arXiv) — Reinforcing driving policies with 3DGS environment for high-fidelity RL training

- **GS-ID** [no arXiv ID; verify primary source] (ICCV 2025) — GS-based illumination decomposition: real-scene light field factorization for embodied intelligence and autonomous driving - **GSMem** [arXiv:2603.19137](https://arxiv.org/abs/2

- **GS-ID** [no arXiv ID; verify primary source] (ICCV 2025) — GS-based illumination decomposition: real-scene light field factorization for embodied intelligence and autonomous driving - **GSMem** [arXiv:2603.19137](https://arxiv.org/abs/2

- **GSMem** [arXiv:2603.19137](https://arxiv.org/abs/2603.19137) (arXiv) — Effective embodied exploration requires agents to accumulate and retain spatial knowledge over time. However, existing scene representations, such as discrete scene 

- **GSMem** [arXiv:2603.19137](https://arxiv.org/abs/2603.19137) (arXiv) — Effective embodied exploration requires agents to accumulate and retain spatial knowledge over time. However, existing scene representations, such as discrete scene 

- **GS-Playground** [arXiv:2604.25459](https://arxiv.org/abs/2604.25459) (RSS 2026) — Batch 3DGS + parallel physics at 10^4 FPS for robot learning

- **GS-Playground** [arXiv:2604.25459](https://arxiv.org/abs/2604.25459) (RSS 2026) — Batch 3DGS + parallel physics at 10^4 FPS for robot learning

- **GS-PRM** [arXiv:2610.19541](https://arxiv.org/abs/2610.19541) (arXiv 2026) — Gaussian visual roadmap for image-goal navigation; 3DGS-rendered milestones and local connections form a probabilistic roadmap a local policy can follow

- **GS-PRM** [arXiv:2610.19541](https://arxiv.org/abs/2610.19541) (arXiv 2026) — Gaussian visual roadmap for image-goal navigation; 3DGS-rendered milestones and local connections form a probabilistic roadmap a local policy can follow

- **GS-Surrogate** [arXiv:2604.06358](https://arxiv.org/abs/2604.06358) (arXiv) — Deformable GS surrogate for ensemble simulation exploration

- **GS-Surrogate** [arXiv:2604.06358](https://arxiv.org/abs/2604.06358) (arXiv) — Deformable GS surrogate for ensemble simulation exploration

- **GS-World** [no arXiv ID; verify primary source] (arXiv 2025) — World model generative simulation + Engine-driven Sim2Real VLA: 3DGS as differentiable simulation engine for robot skill learning - **LeGS** [arXiv:2605.00408](https://arxiv

- **GS-World** [no arXiv ID; verify primary source] (arXiv 2025) — World model generative simulation + Engine-driven Sim2Real VLA: 3DGS as differentiable simulation engine for robot skill learning - **LeGS** [arXiv:2605.00408](https://arxiv

- **LeGS** [arXiv:2605.00408](https://arxiv.org/abs/2605.00408) (arXiv 2026) — Loco-manipulation via Embodied Gaussian Splatting; mesh foreground (robot/objects) + 3DGS background; procedural motion primitive generator; VLA policy trained p

- **LeGS** [arXiv:2605.00408](https://arxiv.org/abs/2605.00408) (arXiv 2026) — Loco-manipulation via Embodied Gaussian Splatting; mesh foreground (robot/objects) + 3DGS background; procedural motion primitive generator; VLA policy trained p

- **LiftNav** [arXiv:2605.31376](https://arxiv.org/abs/2605.31376) (arXiv 2026) — Hybrid navigation framework built on GSFusion's TSDF+GS dual map, augmented with a real-time pipeline of YOLO-based detection, TSDF-based 3D lifting, and B-sp

- **LiftNav** [arXiv:2605.31376](https://arxiv.org/abs/2605.31376) (arXiv 2026) — Hybrid navigation framework built on GSFusion's TSDF+GS dual map, augmented with a real-time pipeline of YOLO-based detection, TSDF-based 3D lifting, and B-sp

- **LOGOS** [arXiv:2606.21527](https://arxiv.org/abs/2606.21527) (arXiv 2026) — To overcome these bottlenecks, we propose LOGOS, a LiDAR-only unified tiny obstacle segmentation system, which models the road surface as a continuous mixture o

- **LOGOS** [arXiv:2606.21527](https://arxiv.org/abs/2606.21527) (arXiv 2026) — To overcome these bottlenecks, we propose LOGOS, a LiDAR-only unified tiny obstacle segmentation system, which models the road surface as a continuous mixture o

- **ManiGaussian** [arXiv:2403.08321](https://arxiv.org/abs/2403.08321) (ECCV 2024) — Dynamic Gaussian Splatting for multi-task robotic manipulation — Gaussian world model predicts optimal actions

- **ManiGaussian** [arXiv:2403.08321](https://arxiv.org/abs/2403.08321) (ECCV 2024) — Dynamic Gaussian Splatting for multi-task robotic manipulation — Gaussian world model predicts optimal actions

- **ManiSplat** [arXiv:2606.10645](https://arxiv.org/abs/2606.10645) (arXiv 2026) — To address these challenges, we introduce ManiSplat, a unified framework that reconstructs controllable and decoupled Gaussian digital twins directly from m

- **ManiSplat** [arXiv:2606.10645](https://arxiv.org/abs/2606.10645) (arXiv 2026) — To address these challenges, we introduce ManiSplat, a unified framework that reconstructs controllable and decoupled Gaussian digital twins directly from m

- **NovaSplat** [arXiv:2610.13309](https://arxiv.org/abs/2610.13309) (arXiv 2026) — Model-free 6D object pose estimation pipeline that trains a 3DGS model from a short onboarding video to serve as both template source and differentiable ref

- **NovaSplat** [arXiv:2610.13309](https://arxiv.org/abs/2610.13309) (arXiv 2026) — Model-free 6D object pose estimation pipeline that trains a 3DGS model from a short onboarding video to serve as both template source and differentiable ref

- **PARTICULATE** [no arXiv ID; verify primary source] (CVPR 2026) — Feed-forward 3D object articulation from static mesh; auto-infer movable structure including parts, connections, and motion axes; Oxford / Cambridge / NTU - **PG-Pose** [a

- **PARTICULATE** [no arXiv ID; verify primary source] (CVPR 2026) — Feed-forward 3D object articulation from static mesh; auto-infer movable structure including parts, connections, and motion axes; Oxford / Cambridge / NTU - **PG-Pose** [a

- **PG-Pose** [arXiv:2609.07231](https://arxiv.org/abs/2609.07231) (arXiv 2026) — PG-Pose: geometry-aware framework combining Planar-based Gaussian Splatting (PGS) reconstruction with geometry-driven pose optimization; generalizable 6D pose

- **PG-Pose** [arXiv:2609.07231](https://arxiv.org/abs/2609.07231) (arXiv 2026) — PG-Pose: geometry-aware framework combining Planar-based Gaussian Splatting (PGS) reconstruction with geometry-driven pose optimization; generalizable 6D pose

- **PolyMerge** [arXiv:2606.16232](https://arxiv.org/abs/2606.16232) (arXiv 2026) — To address these limitations, we propose PolyMerge to convert a large, photorealistic 3D Gaussian Splatting (3DGS) model of a scene into a lightweight repre

- **PolyMerge** [arXiv:2606.16232](https://arxiv.org/abs/2606.16232) (arXiv 2026) — To address these limitations, we propose PolyMerge to convert a large, photorealistic 3D Gaussian Splatting (3DGS) model of a scene into a lightweight repre

- **PrecisionReach-GS** [arXiv:2610.35626](https://arxiv.org/abs/2610.35626) (arXiv 2026) — 3DGS-based data generation pipeline for wrist-camera VLA adaptation; reconstructs the real workspace with 3DGS, renders robot-aligned wrist-camera o

- **PrecisionReach-GS** [arXiv:2610.35626](https://arxiv.org/abs/2610.35626) (arXiv 2026) — 3DGS-based data generation pipeline for wrist-camera VLA adaptation; reconstructs the real workspace with 3DGS, renders robot-aligned wrist-camera o

- **ProbSplat** [arXiv:2608.13143](https://arxiv.org/abs/2608.13143) (arXiv 2026) — This paper presents ProbSplat, a Compute-in-Memory (CIM)-inspired architecture based on programmable and energy efficient floating-gate inverter columns for

- **ProbSplat** [arXiv:2608.13143](https://arxiv.org/abs/2608.13143) (arXiv 2026) — This paper presents ProbSplat, a Compute-in-Memory (CIM)-inspired architecture based on programmable and energy efficient floating-gate inverter columns for

- **Real2Sem2Real** [arXiv:2610.28880](https://arxiv.org/abs/2610.28880) (arXiv 2026) — Real-to-semantic-to-real pipeline turning one monocular RGB human demonstration into thousands of robot demonstrations; tracks object SE(3) poses with D

- **Real2Sem2Real** [arXiv:2610.28880](https://arxiv.org/abs/2610.28880) (arXiv 2026) — Real-to-semantic-to-real pipeline turning one monocular RGB human demonstration into thousands of robot demonstrations; tracks object SE(3) poses with D

- **Real2Sim** [arXiv:2605.13591](https://arxiv.org/abs/2605.13591) (arXiv 2026) — 4DGS + differentiable MPM solver for physics-aware autonomous driving simulation

- **Real2Sim** [arXiv:2605.13591](https://arxiv.org/abs/2605.13591) (arXiv 2026) — 4DGS + differentiable MPM solver for physics-aware autonomous driving simulation

- **RefGlass-GS** [arXiv:2606.28826](https://arxiv.org/abs/2606.28826) (arXiv 2026) — Therefore, we propose RefGlass-GS, a fusion framework that enables end-to-end UAV-based photorealistic, semantic, and interactive digitization of reflecti

- **RefGlass-GS** [arXiv:2606.28826](https://arxiv.org/abs/2606.28826) (arXiv 2026) — Therefore, we propose RefGlass-GS, a fusion framework that enables end-to-end UAV-based photorealistic, semantic, and interactive digitization of reflecti

- **RIDE** [arXiv:2609.11079](https://arxiv.org/abs/2609.11079) (arXiv 2026) — Relocalization-informed depth estimation with 3DGS; render-match-PnP relocalization supports dense depth estimation

- **RIDE** [arXiv:2609.11079](https://arxiv.org/abs/2609.11079) (arXiv 2026) — Relocalization-informed depth estimation with 3DGS; render-match-PnP relocalization supports dense depth estimation

- **SA-ResGS** [arXiv:2601.03024](https://arxiv.org/abs/2601.03024) (ECCV 2026) — Self-augmented residual 3DGS for next-best-view selection; SA-Points via triangulation for coverage estimation; first residual learning for 3DGS to improve gr

- **SA-ResGS** [arXiv:2601.03024](https://arxiv.org/abs/2601.03024) (ECCV 2026) — Self-augmented residual 3DGS for next-best-view selection; SA-Points via triangulation for coverage estimation; first residual learning for 3DGS to improve gr

- **SceneSmith** [arXiv:2602.09153](https://arxiv.org/abs/2602.09153) (ICML 2026 Spotlight) — Hierarchical agentic framework for simulation-ready indoor scene generation from natural language; designer-critic-orchestrator VLM agents; text-t

- **SceneSmith** [arXiv:2602.09153](https://arxiv.org/abs/2602.09153) (ICML 2026 Spotlight) — Hierarchical agentic framework for simulation-ready indoor scene generation from natural language; designer-critic-orchestrator VLM agents; text-t

- **SemSafe-3DGS** [arXiv:2609.19330](https://arxiv.org/abs/2609.19330) (arXiv 2026) — Semantic risk-aware safe-active perception navigation framework in attributed 3D Gaussian maps.

- **SemSafe-3DGS** [arXiv:2609.19330](https://arxiv.org/abs/2609.19330) (arXiv 2026) — Semantic risk-aware safe-active perception navigation framework in attributed 3D Gaussian maps.

- **Splat2Act** [arXiv:2610.13945](https://arxiv.org/abs/2610.13945) (arXiv 2026) — Automated real-to-sim-to-real pipeline that personalizes a pretrained VLA policy to a specific scene and object set using only a handheld phone scan and obj

- **Splat2Act** [arXiv:2610.13945](https://arxiv.org/abs/2610.13945) (arXiv 2026) — Automated real-to-sim-to-real pipeline that personalizes a pretrained VLA policy to a specific scene and object set using only a handheld phone scan and obj

- **Splat-CBF** [arXiv:2609.23100](https://arxiv.org/abs/2609.23100) (arXiv 2026) — Risk-aware control barrier function for safe next-best-view control in 3D Gaussian-Splat maps.

- **Splat-CBF** [arXiv:2609.23100](https://arxiv.org/abs/2609.23100) (arXiv 2026) — Risk-aware control barrier function for safe next-best-view control in 3D Gaussian-Splat maps.

- **SplatCtrl** [arXiv:2607.08948](https://arxiv.org/abs/2607.08948) (ICRA 2026) — Perception-action coupling via GS + reactive robot control; isotropic Gaussian → continuous SDF → control barrier functions; collision-free manipulation

- **SplatCtrl** [arXiv:2607.08948](https://arxiv.org/abs/2607.08948) (ICRA 2026) — Perception-action coupling via GS + reactive robot control; isotropic Gaussian → continuous SDF → control barrier functions; collision-free manipulation

- **Splat-Nav** [arXiv:2403.02751](https://arxiv.org/abs/2403.02751) (CVPR 2025) — GS-based navigation with Gaussian-anchored topological maps

- **Splat-Nav** [arXiv:2403.02751](https://arxiv.org/abs/2403.02751) (CVPR 2025) — GS-based navigation with Gaussian-anchored topological maps

- **SplatSim** [arXiv:2409.10161](https://arxiv.org/abs/2409.10161) (CVPR 2025) — GS-based sim-to-real transfer for robotic manipulation with photorealistic rendering

- **SplatSim** [arXiv:2409.10161](https://arxiv.org/abs/2409.10161) (CVPR 2025) — GS-based sim-to-real transfer for robotic manipulation with photorealistic rendering

- **SpotlessGS** [arXiv:2608.14713](https://arxiv.org/abs/2608.14713) (arXiv 2026) — Robots operating in dark or poorly lit environments rely on onboard lights, which often produce uneven illumination that degrades downstream perception tas

- **SpotlessGS** [arXiv:2608.14713](https://arxiv.org/abs/2608.14713) (arXiv 2026) — Robots operating in dark or poorly lit environments rely on onboard lights, which often produce uneven illumination that degrades downstream perception tas

- **TAIL-Safe** [arXiv:2605.01195](https://arxiv.org/abs/2605.01195) (arXiv) — Safety monitoring for IL policies using 3DGS digital twin

- **TAIL-Safe** [arXiv:2605.01195](https://arxiv.org/abs/2605.01195) (arXiv) — Safety monitoring for IL policies using 3DGS digital twin

- **VR-Robo** [arXiv:2502.01536](https://arxiv.org/abs/2502.01536) (RAL 2025) — Recent success in legged robot locomotion is attributed to the integration of reinforcement learning and physical simulators. However, these policies often enco

- **VR-Robo** [arXiv:2502.01536](https://arxiv.org/abs/2502.01536) (RAL 2025) — Recent success in legged robot locomotion is attributed to the integration of reinforcement learning and physical simulators. However, these policies often enco

- **2DGS-SLAM** [no arXiv ID; verify primary source] (TRO 2026) — Globally consistent RGB-D SLAM with 2DGS; loop closure + global optimization - **Anchor3R** [arXiv:2606.05035](https://arxiv.org/abs/2606.05035) (arXiv) — Streaming 3D recons

- **2DGS-SLAM** [no arXiv ID; verify primary source] (TRO 2026) — Globally consistent RGB-D SLAM with 2DGS; loop closure + global optimization - **Anchor3R** [arXiv:2606.05035](https://arxiv.org/abs/2606.05035) (arXiv) — Streaming 3D recons

- **Anchor3R** [arXiv:2606.05035](https://arxiv.org/abs/2606.05035) (arXiv) — Streaming 3D reconstruction with current-centric local prediction; transient anchor + loop closure

- **Anchor3R** [arXiv:2606.05035](https://arxiv.org/abs/2606.05035) (arXiv) — Streaming 3D reconstruction with current-centric local prediction; transient anchor + loop closure

- **Anythingreality** [arXiv:2607.09260](https://arxiv.org/abs/2607.09260) (arXiv) — Robust online GS SLAM for VR scene exploration + VLM interaction; ORB-SLAM3-based pose + real-time Gaussian reconstruction; +14.5% PSNR, 88% VLM recognitio

- **Anythingreality** [arXiv:2607.09260](https://arxiv.org/abs/2607.09260) (arXiv) — Robust online GS SLAM for VR scene exploration + VLM interaction; ORB-SLAM3-based pose + real-time Gaussian reconstruction; +14.5% PSNR, 88% VLM recognitio

- **CGS-SLAM** [no arXiv ID; verify primary source] (IROS 2025) — Compact voxel-based 3DGS for dense visual SLAM - **CoMapGS** [arXiv:2503.20998](https://arxiv.org/abs/2503.20998) (CVPR 2025) — Covisibility map-based Gaussian Splatting for 

- **CGS-SLAM** [no arXiv ID; verify primary source] (IROS 2025) — Compact voxel-based 3DGS for dense visual SLAM - **CoMapGS** [arXiv:2503.20998](https://arxiv.org/abs/2503.20998) (CVPR 2025) — Covisibility map-based Gaussian Splatting for 

- **CoMapGS** [arXiv:2503.20998](https://arxiv.org/abs/2503.20998) (CVPR 2025) — Covisibility map-based Gaussian Splatting for sparse-view novel view synthesis and rendering

- **CoMapGS** [arXiv:2503.20998](https://arxiv.org/abs/2503.20998) (CVPR 2025) — Covisibility map-based Gaussian Splatting for sparse-view novel view synthesis and rendering

- **Cube-Splat** [arXiv:2609.21347](https://arxiv.org/abs/2609.21347) (arXiv 2026) — First panoramic GS-SLAM framework that factorizes each 360° frame into a cubemap of four fixed-orientation virtual pinhole views sharing a single optical c

- **Cube-Splat** [arXiv:2609.21347](https://arxiv.org/abs/2609.21347) (arXiv 2026) — First panoramic GS-SLAM framework that factorizes each 360° frame into a cubemap of four fixed-orientation virtual pinhole views sharing a single optical c

- **DG-SLAM** [arXiv:2411.08373](https://arxiv.org/abs/2411.08373) (NeurIPS 2024) — Dynamic Gaussian SLAM with hybrid pose optimization for dynamic environments

- **DG-SLAM** [arXiv:2411.08373](https://arxiv.org/abs/2411.08373) (NeurIPS 2024) — Dynamic Gaussian SLAM with hybrid pose optimization for dynamic environments

- **DL-SLAM** [no arXiv ID; verify primary source] (arXiv) — Dual-level probability framework for dynamic environment GS-SLAM; semantic+geometric pixel/object-level dynamic probability; artifact-free static map + semantic map - **EliGSiR** 

- **DL-SLAM** [no arXiv ID; verify primary source] (arXiv) — Dual-level probability framework for dynamic environment GS-SLAM; semantic+geometric pixel/object-level dynamic probability; artifact-free static map + semantic map - **EliGSiR** 

- **EliGSiR** [arXiv:2609.20348](https://arxiv.org/abs/2609.20348) (arXiv 2026) — Continual Gaussian mapper that controls how the available optimization budget is used as the reconstruction evolves

- **EliGSiR** [arXiv:2609.20348](https://arxiv.org/abs/2609.20348) (arXiv 2026) — Continual Gaussian mapper that controls how the available optimization budget is used as the reconstruction evolves

- **EvFlow-GS** [arXiv:2604.22183](https://arxiv.org/abs/2604.22183) (ICME 2026) — Event camera + optical flow for motion blur handling

- **EvFlow-GS** [arXiv:2604.22183](https://arxiv.org/abs/2604.22183) (ICME 2026) — Event camera + optical flow for motion blur handling

- **Flow4DGS-SLAM** [arXiv:2604.22339](https://arxiv.org/abs/2604.22339) (CVPR 2026) — Optical flow-guided 4D Gaussian SLAM for dynamic scenes; category-agnostic motion mask; GMM temporal opacity/rotation

- **Flow4DGS-SLAM** [arXiv:2604.22339](https://arxiv.org/abs/2604.22339) (CVPR 2026) — Optical flow-guided 4D Gaussian SLAM for dynamic scenes; category-agnostic motion mask; GMM temporal opacity/rotation

- **Gaussian Splatting SLAM** [arXiv:2312.06741](https://arxiv.org/abs/2312.06741) (CVPR 2024 (Highlight)) — First real-time monocular 3DGS SLAM

- **Gaussian Splatting SLAM** [arXiv:2312.06741](https://arxiv.org/abs/2312.06741) (CVPR 2024 (Highlight)) — First real-time monocular 3DGS SLAM

- **GAVIS** [arXiv:2605.30342](https://arxiv.org/abs/2605.30342) (CVPR) — Uncertainty-driven 3DGS active mapping via anisotropic visibility field; Bayesian Network rasterizer

- **GAVIS** [arXiv:2605.30342](https://arxiv.org/abs/2605.30342) (CVPR) — Uncertainty-driven 3DGS active mapping via anisotropic visibility field; Bayesian Network rasterizer

- **GeoGS-SLAM** [arXiv:2607.07452](https://arxiv.org/abs/2607.07452) (arXiv) — Geometry-Only Gaussian Splatting for dense monocular SLAM; removes appearance modeling reducing per-primitive params 80%+; SOTA online mapping + geometry qualit

- **GeoGS-SLAM** [arXiv:2607.07452](https://arxiv.org/abs/2607.07452) (arXiv) — Geometry-Only Gaussian Splatting for dense monocular SLAM; removes appearance modeling reducing per-primitive params 80%+; SOTA online mapping + geometry qualit

- **GeoGS-SLAM (Geometric Priors)** [arXiv:2607.11184](https://arxiv.org/abs/2607.11184) (arXiv) — Online monocular dense reconstruction combining a 3DGS scene map with learned geometric priors from a feed-forward visual geometry model; Gau

- **GeoGS-SLAM (Geometric Priors)** [arXiv:2607.11184](https://arxiv.org/abs/2607.11184) (arXiv) — Online monocular dense reconstruction combining a 3DGS scene map with learned geometric priors from a feed-forward visual geometry model; Gau

- **GGD-SLAM** [arXiv:2604.12837](https://arxiv.org/abs/2604.12837) (ICRA 2026) — Generalizable motion model for monocular dynamic 3DGS SLAM; FIFO sequential attention; no semantic labels/depth required

- **GGD-SLAM** [arXiv:2604.12837](https://arxiv.org/abs/2604.12837) (ICRA 2026) — Generalizable motion model for monocular dynamic 3DGS SLAM; FIFO sequential attention; no semantic labels/depth required

- **GLAM-SLAM** [arXiv:2607.21416](https://arxiv.org/abs/2607.21416) (IROS 2026) — Large-scale outdoor decoupled GS SLAM; feature tracking front-end + sparse anchor grid; 15% better on KITTI/Oxford

- **GLAM-SLAM** [arXiv:2607.21416](https://arxiv.org/abs/2607.21416) (IROS 2026) — Large-scale outdoor decoupled GS SLAM; feature tracking front-end + sparse anchor grid; 15% better on KITTI/Oxford

- **GOFusion** [arXiv:2610.25261](https://arxiv.org/abs/2610.25261) (arXiv preprint) — Gaussian-occupancy submap fusion for visual mapping and information-driven navigation, with active-submap optimization.

- **GOFusion** [arXiv:2610.25261](https://arxiv.org/abs/2610.25261) (arXiv preprint) — Gaussian-occupancy submap fusion for visual mapping and information-driven navigation, with active-submap optimization.

- **GS-Map-SLAM** [no arXiv ID; verify primary source] (CVPR) — Gaussian-based dense mapping for SLAM with view-dependent quality; real-time dense mapping via GS optimization - **HyperLoop** [arXiv:2610.19742](https://arxiv.org/abs/2610.197

- **GS-Map-SLAM** [no arXiv ID; verify primary source] (CVPR) — Gaussian-based dense mapping for SLAM with view-dependent quality; real-time dense mapping via GS optimization - **HyperLoop** [arXiv:2610.19742](https://arxiv.org/abs/2610.197

- **HyperLoop** [arXiv:2610.19742](https://arxiv.org/abs/2610.19742) (arXiv 2026) — Hyperspectral-assisted loop closure detection in 3DGS SLAM for planetary exploration; counters perceptual aliasing on repetitive featureless terrain

- **HyperLoop** [arXiv:2610.19742](https://arxiv.org/abs/2610.19742) (arXiv 2026) — Hyperspectral-assisted loop closure detection in 3DGS SLAM for planetary exploration; counters perceptual aliasing on repetitive featureless terrain

- **ImprovedVBGS** [arXiv:2607.15542](https://arxiv.org/abs/2607.15542) (arXiv 2026) — Real-time continual Variational Bayes GS; improved CAVI with per-frame subset updates instead of full-observation iterations; enables on-the-fly SLAM wit

- **ImprovedVBGS** [arXiv:2607.15542](https://arxiv.org/abs/2607.15542) (arXiv 2026) — Real-time continual Variational Bayes GS; improved CAVI with per-frame subset updates instead of full-observation iterations; enables on-the-fly SLAM wit

- **KFS4LAM** [arXiv:2610.22505](https://arxiv.org/abs/2610.22505) (arXiv 2026) — Adaptive keyframe selection with structure-aware mapping for endoscopic GS SLAM; uncertainty-driven retention beats fixed temporal intervals

- **KFS4LAM** [arXiv:2610.22505](https://arxiv.org/abs/2610.22505) (arXiv 2026) — Adaptive keyframe selection with structure-aware mapping for endoscopic GS SLAM; uncertainty-driven retention beats fixed temporal intervals

- **LightSplat** [arXiv:2609.07274](https://arxiv.org/abs/2609.07274) (arXiv 2026) — Real-time high-fidelity 3DGS SLAM with loop closure; meets real-time constraints with loop closure support

- **LightSplat** [arXiv:2609.07274](https://arxiv.org/abs/2609.07274) (arXiv 2026) — Real-time high-fidelity 3DGS SLAM with loop closure; meets real-time constraints with loop closure support

- **LingGS** [arXiv:2610.36853](https://arxiv.org/abs/2610.36853) (arXiv 2026) — Gaussian SLAM decoupling feed-forward geometry prediction from camera tracking; combines learned geometric priors with explicit geometric camera tracking for h

- **LingGS** [arXiv:2610.36853](https://arxiv.org/abs/2610.36853) (arXiv 2026) — Gaussian SLAM decoupling feed-forward geometry prediction from camera tracking; combines learned geometric priors with explicit geometric camera tracking for h

- **LiteLoc** [arXiv:2605.17777](https://arxiv.org/abs/2605.17777) (IEEE/CAA JAS 2026) — Color-free decoupled 3DGS feature field for visual localization: 94% storage reduction + 5% match condensing

- **LiteLoc** [arXiv:2605.17777](https://arxiv.org/abs/2605.17777) (IEEE/CAA JAS 2026) — Color-free decoupled 3DGS feature field for visual localization: 94% storage reduction + 5% match condensing

- **MAGICIAN** [arXiv:2603.22650](https://arxiv.org/abs/2603.22650) (CVPR 2026 Oral) — Active mapping with imagined Gaussians + beam search

- **MAGICIAN** [arXiv:2603.22650](https://arxiv.org/abs/2603.22650) (CVPR 2026 Oral) — Active mapping with imagined Gaussians + beam search

- **MAGS-SLAM** [arXiv:2605.10760](https://arxiv.org/abs/2605.10760) (arXiv) — Collaborative photorealistic 3D reconstruction from multiple agents enables rapid large-scale scene capture for virtual production and cooperative multi-robot ex

- **MAGS-SLAM** [arXiv:2605.10760](https://arxiv.org/abs/2605.10760) (arXiv) — Collaborative photorealistic 3D reconstruction from multiple agents enables rapid large-scale scene capture for virtual production and cooperative multi-robot ex

- **MMD-SLAM** [arXiv:2606.19874](https://arxiv.org/abs/2606.19874) (arXiv 2026) — To address these limitations, we propose MMD-SLAM, a structure-enhanced Visual SLAM framework that leverages the Atlanta World (AW) assumption to guide a Mul

- **MMD-SLAM** [arXiv:2606.19874](https://arxiv.org/abs/2606.19874) (arXiv 2026) — To address these limitations, we propose MMD-SLAM, a structure-enhanced Visual SLAM framework that leverages the Atlanta World (AW) assumption to guide a Mul

- **MonoEM-GS** [arXiv:2604.10593](https://arxiv.org/abs/2604.10593) (arXiv) — Monocular expectation-maximization GS SLAM

- **MonoEM-GS** [arXiv:2604.10593](https://arxiv.org/abs/2604.10593) (arXiv) — Monocular expectation-maximization GS SLAM

- **MoonSplat** [arXiv:2606.17935](https://arxiv.org/abs/2606.17935) (arXiv 2026) — Our code and data are available at https://github.com/TrickyGo/MoonSplat

- **MoonSplat** [arXiv:2606.17935](https://arxiv.org/abs/2606.17935) (arXiv 2026) — Our code and data are available at https://github.com/TrickyGo/MoonSplat

- **MoPe** [arXiv:2606.29237](https://arxiv.org/abs/2606.29237) (arXiv 2026) — We realize this principle in MoPe, a memory-aware uncertainty filter for monocular Gaussian mapping

- **MoPe** [arXiv:2606.29237](https://arxiv.org/abs/2606.29237) (arXiv 2026) — We realize this principle in MoPe, a memory-aware uncertainty filter for monocular Gaussian mapping

- **MyGO-Splat** [arXiv:2606.29738](https://arxiv.org/abs/2606.29738) (arXiv 2026) — Closed-loop Gaussian SLAM framework that analytically rasterizes Gaussian primitives into pixel-wise depth and surface normals, allowing the map to activel

- **MyGO-Splat** [arXiv:2606.29738](https://arxiv.org/abs/2606.29738) (arXiv 2026) — Closed-loop Gaussian SLAM framework that analytically rasterizes Gaussian primitives into pixel-wise depth and surface normals, allowing the map to activel

- **PanoGS-SLAM** [arXiv:2609.17387](https://arxiv.org/abs/2609.17387) (arXiv 2026) — First panoramic dense SLAM on 3DGS; spherical-domain differentiable rendering, sphere-consistent photometric loss, depth-guided Gaussian init

- **PanoGS-SLAM** [arXiv:2609.17387](https://arxiv.org/abs/2609.17387) (arXiv 2026) — First panoramic dense SLAM on 3DGS; spherical-domain differentiable rendering, sphere-consistent photometric loss, depth-guided Gaussian init

- **Photo-SLAM** [arXiv:2311.16728](https://arxiv.org/abs/2311.16728) (CVPR 2024) — Hyper primitives map with explicit geometric features for localization + implicit photometric features; Gaussian-Pyramid training; runs on Jetson AGX Orin

- **Photo-SLAM** [arXiv:2311.16728](https://arxiv.org/abs/2311.16728) (CVPR 2024) — Hyper primitives map with explicit geometric features for localization + implicit photometric features; Gaussian-Pyramid training; runs on Jetson AGX Orin

- **Pocket-SLAM** [arXiv:2606.24796](https://arxiv.org/abs/2606.24796) (arXiv 2026) — Rendering-area-aware pruning strategy that selectively removes Gaussians based on their contribution to the effective rendering area, rather than solely r

- **Pocket-SLAM** [arXiv:2606.24796](https://arxiv.org/abs/2606.24796) (arXiv 2026) — Rendering-area-aware pruning strategy that selectively removes Gaussians based on their contribution to the effective rendering area, rather than solely r

- **RawSLAM** [arXiv:2609.20589](https://arxiv.org/abs/2609.20589) (arXiv 2026) — To the best of our knowledge, the first online Gaussian SLAM framework that tracks and maps directly on single-exposure 16-bit linear HDR imagery

- **RawSLAM** [arXiv:2609.20589](https://arxiv.org/abs/2609.20589) (arXiv 2026) — To the best of our knowledge, the first online Gaussian SLAM framework that tracks and maps directly on single-exposure 16-bit linear HDR imagery

- **Real-Time LiDAR GS-SLAM** [no arXiv ID; verify primary source] (arXiv) — First real-time LiDAR Gaussian Splatting SLAM; tight G-ICP + spherical rasterization dense mapping; Newer College F-score 86.78%@>20FPS - **RoMo-GS** [arXiv:2610.2

- **Real-Time LiDAR GS-SLAM** [no arXiv ID; verify primary source] (arXiv) — First real-time LiDAR Gaussian Splatting SLAM; tight G-ICP + spherical rasterization dense mapping; Newer College F-score 86.78%@>20FPS - **RoMo-GS** [arXiv:2610.2

- **RoMo-GS** [arXiv:2610.24605](https://arxiv.org/abs/2610.24605) (arXiv preprint) — Abstract context: Online 3D Gaussian Splatting (3DGS) is emerging as a practical way to build a representation for robotics applications.

- **RoMo-GS** [arXiv:2610.24605](https://arxiv.org/abs/2610.24605) (arXiv preprint) — Abstract context: Online 3D Gaussian Splatting (3DGS) is emerging as a practical way to build a representation for robotics applications.

- **RoSe-SLAM** [arXiv:2608.29003](https://arxiv.org/abs/2608.29003) (arXiv 2026) — Robust Semantic-aware Gaussian Splatting SLAM (RoSe-SLAM), to address the dynamic challenge by a holistic semantic scene understanding from uncalibrated mon

- **RoSe-SLAM** [arXiv:2608.29003](https://arxiv.org/abs/2608.29003) (arXiv 2026) — Robust Semantic-aware Gaussian Splatting SLAM (RoSe-SLAM), to address the dynamic challenge by a holistic semantic scene understanding from uncalibrated mon

- **S3PO-GS** [no arXiv ID; verify primary source] (ICCV 2025) — Global scale-consistent outdoor monocular 3DGS SLAM - **SCOUT-SLAM** [arXiv:2609.14634](https://arxiv.org/abs/2609.14634) (arXiv 2026) — Structurally-coupled dual-uncertainty 

- **S3PO-GS** [no arXiv ID; verify primary source] (ICCV 2025) — Global scale-consistent outdoor monocular 3DGS SLAM - **SCOUT-SLAM** [arXiv:2609.14634](https://arxiv.org/abs/2609.14634) (arXiv 2026) — Structurally-coupled dual-uncertainty 

- **SCOUT-SLAM** [arXiv:2609.14634](https://arxiv.org/abs/2609.14634) (arXiv 2026) — Structurally-coupled dual-uncertainty 3DGS SLAM; shared-base LoAd tracking uncertainty independent of reconstruction quality

- **SCOUT-SLAM** [arXiv:2609.14634](https://arxiv.org/abs/2609.14634) (arXiv 2026) — Structurally-coupled dual-uncertainty 3DGS SLAM; shared-base LoAd tracking uncertainty independent of reconstruction quality

- **SG2SLAM** [arXiv:2610.36312](https://arxiv.org/abs/2610.36312) (arXiv 2026) — Spatial geometry-informed 3DGS SLAM for monocular outdoor navigation; couples streaming spatial geometry with a ray-conditioned 3DGS kernel

- **SG2SLAM** [arXiv:2610.36312](https://arxiv.org/abs/2610.36312) (arXiv 2026) — Spatial geometry-informed 3DGS SLAM for monocular outdoor navigation; couples streaming spatial geometry with a ray-conditioned 3DGS kernel

- **SplaTAM** [arXiv:2312.02126](https://arxiv.org/abs/2312.02126) (CVPR 2024) — First real-time GS-SLAM: online incremental Gaussians with silhouette mask and rendering-based pose optimization

- **SplaTAM** [arXiv:2312.02126](https://arxiv.org/abs/2312.02126) (CVPR 2024) — First real-time GS-SLAM: online incremental Gaussians with silhouette mask and rendering-based pose optimization

- **SplatlessDF** [arXiv:2606.13990](https://arxiv.org/abs/2606.13990) (arXiv 2026) — In this paper, building on this principle, we introduce SplatlessDF, a continuous distance field (DF) mapping framework that uses anisotropic Gaussian ele

- **SplatlessDF** [arXiv:2606.13990](https://arxiv.org/abs/2606.13990) (arXiv 2026) — In this paper, building on this principle, we introduce SplatlessDF, a continuous distance field (DF) mapping framework that uses anisotropic Gaussian ele

- **SplatLoc** [arXiv:2409.14067](https://arxiv.org/abs/2409.14067) (CVPR 2025) — GS-based visual localization with Gaussian-anchored map representation

- **SplatLoc** [arXiv:2409.14067](https://arxiv.org/abs/2409.14067) (CVPR 2025) — GS-based visual localization with Gaussian-anchored map representation

- **SplitGS-Loc** [arXiv:2605.07351](https://arxiv.org/abs/2605.07351) (arXiv 2026) — Mixture-of-Gaussians splitting to disambiguate 2D-3D correspondences in GS feature fields

- **SplitGS-Loc** [arXiv:2605.07351](https://arxiv.org/abs/2605.07351) (arXiv 2026) — Mixture-of-Gaussians splitting to disambiguate 2D-3D correspondences in GS feature fields

- **Structured-Li-GS** [arXiv:2606.27509](https://arxiv.org/abs/2606.27509) (arXiv 2026) — In this study, we develop a Structured framework for Gaussian Splatting (3DGS) with LiDAR integration (Structured-Li-GS)

- **Structured-Li-GS** [arXiv:2606.27509](https://arxiv.org/abs/2606.27509) (arXiv 2026) — In this study, we develop a Structured framework for Gaussian Splatting (3DGS) with LiDAR integration (Structured-Li-GS)

- **Track2Map** [arXiv:2607.08408](https://arxiv.org/abs/2607.08408) (MICCAI 2026) — Online deformable GS SLAM for robotic surgery; track-anchored deformation initialization; joint camera + scene optimization from surgical video

- **Track2Map** [arXiv:2607.08408](https://arxiv.org/abs/2607.08408) (MICCAI 2026) — Online deformable GS SLAM for robotic surgery; track-anchored deformation initialization; joint camera + scene optimization from surgical video

- **Triangle Splatting SLAM** [arXiv:2605.31419](https://arxiv.org/abs/2605.31419) (arXiv 2026) — First dense RGB-D SLAM with differentiable triangles as 3D map representation; Delaunay triangulation converts triangle soup to connected mesh

- **Triangle Splatting SLAM** [arXiv:2605.31419](https://arxiv.org/abs/2605.31419) (arXiv 2026) — First dense RGB-D SLAM with differentiable triangles as 3D map representation; Delaunay triangulation converts triangle soup to connected mesh

- **ULF-Loc** [arXiv:2605.04730](https://arxiv.org/abs/2605.04730) (CVPR 2026 Highlight) — Unbiased landmark feature for robust visual localization with 3DGS; geometry-weighted aggregation replaces alpha-compositing features

- **ULF-Loc** [arXiv:2605.04730](https://arxiv.org/abs/2605.04730) (CVPR 2026 Highlight) — Unbiased landmark feature for robust visual localization with 3DGS; geometry-weighted aggregation replaces alpha-compositing features

- **VGGT-GS SLAM** [arXiv:2609.19628](https://arxiv.org/abs/2609.19628) (arXiv 2026) — Monocular 3D Gaussian Splatting SLAM system designed for uncalibrated videos

- **VGGT-GS SLAM** [arXiv:2609.19628](https://arxiv.org/abs/2609.19628) (arXiv 2026) — Monocular 3D Gaussian Splatting SLAM system designed for uncalibrated videos

- **VIGS-SLAM** [arXiv:2512.02293](https://arxiv.org/abs/2512.02293) (ECCV 2026) — Visual-inertial GS SLAM with iPhone real-time demo; feature tracking + Gaussian mapping; Docker + TensorRT

- **VIGS-SLAM** [arXiv:2512.02293](https://arxiv.org/abs/2512.02293) (ECCV 2026) — Visual-inertial GS SLAM with iPhone real-time demo; feature tracking + Gaussian mapping; Docker + TensorRT

- **WildGS-SLAM** [arXiv:2504.03886](https://arxiv.org/abs/2504.03886) (CVPR 2025) — Dynamic environment SLAM with uncertainty-aware mapping

- **WildGS-SLAM** [arXiv:2504.03886](https://arxiv.org/abs/2504.03886) (CVPR 2025) — Dynamic environment SLAM with uncertainty-aware mapping

- **3DEditSafe** [arXiv:2605.15398](https://arxiv.org/abs/2605.15398) (arXiv 2026) — First safety-regularized 3D editing framework constraining NSFW semantic propagation; 3D safety regularization + safe semantic projection + residue suppres

- **3DEditSafe** [arXiv:2605.15398](https://arxiv.org/abs/2605.15398) (arXiv 2026) — First safety-regularized 3D editing framework constraining NSFW semantic propagation; 3D safety regularization + safe semantic projection + residue suppres

- **4D-GSW** [arXiv:2605.22342](https://arxiv.org/abs/2605.22342) (arXiv) — Kinematic-aware watermarking via STC metric + HMM-MRF energy for spatio-temporal consistency in 4DGS

- **4D-GSW** [arXiv:2605.22342](https://arxiv.org/abs/2605.22342) (arXiv) — Kinematic-aware watermarking via STC metric + HMM-MRF energy for spatio-temporal consistency in 4DGS

- **BitC-3DGS** [arXiv:2605.29583](https://arxiv.org/abs/2605.29583) (arXiv) — High-capacity 3DGS watermarking via bit-plane complement coding

- **BitC-3DGS** [arXiv:2605.29583](https://arxiv.org/abs/2605.29583) (arXiv) — High-capacity 3DGS watermarking via bit-plane complement coding

- **FlyMark** [arXiv:2610.11364](https://arxiv.org/abs/2610.11364) (arXiv 2026) — Training-free invisible watermarking that writes a keyed message into 3DGS parameters using carrier directions derived from a fruit-fly connectome; extraction

- **FlyMark** [arXiv:2610.11364](https://arxiv.org/abs/2610.11364) (arXiv 2026) — Training-free invisible watermarking that writes a keyed message into 3DGS parameters using carrier directions derived from a fruit-fly connectome; extraction

- **GaussianMarker** [arXiv:2410.23718](https://arxiv.org/abs/2410.23718) (NeurIPS 2024) — Uncertainty-aware watermark embedding + robust extraction for 3DGS copyright protection

- **GaussianMarker** [arXiv:2410.23718](https://arxiv.org/abs/2410.23718) (NeurIPS 2024) — Uncertainty-aware watermark embedding + robust extraction for 3DGS copyright protection

- **GaussTrace** [arXiv:2606.10612](https://arxiv.org/abs/2606.10612) (ICML 2026) — 3DGS provenance analysis via evidence-driven LLM reasoning; directed provenance graphs from Gaussian scene attributes for IP protection and forensics

- **GaussTrace** [arXiv:2606.10612](https://arxiv.org/abs/2606.10612) (ICML 2026) — 3DGS provenance analysis via evidence-driven LLM reasoning; directed provenance graphs from Gaussian scene attributes for IP protection and forensics

- **GeometryCloak** [arXiv:2410.22705](https://arxiv.org/abs/2410.22705) (NeurIPS 2024) — Geometric perturbation copyright watermark embedding into Gaussians preventing TGS-based 3D reconstruction

- **GeometryCloak** [arXiv:2410.22705](https://arxiv.org/abs/2410.22705) (NeurIPS 2024) — Geometric perturbation copyright watermark embedding into Gaussians preventing TGS-based 3D reconstruction

- **GhostSplat** [arXiv:2608.29184](https://arxiv.org/abs/2608.29184) (arXiv 2026) — Input-triggered backdoor that installs such behavior in feed-forward 3DGS

- **GhostSplat** [arXiv:2608.29184](https://arxiv.org/abs/2608.29184) (arXiv 2026) — Input-triggered backdoor that installs such behavior in feed-forward 3DGS

- **GS-Hider** [arXiv:2405.15118](https://arxiv.org/abs/2405.15118) (NeurIPS 2024) — Steganography embedding into Gaussian parameters for 3D message hiding, visually lossless

- **GS-Hider** [arXiv:2405.15118](https://arxiv.org/abs/2405.15118) (NeurIPS 2024) — Steganography embedding into Gaussian parameters for 3D message hiding, visually lossless

- **GuardMarkGS** [arXiv:2605.12919](https://arxiv.org/abs/2605.12919) (arXiv 2026) — 3D Gaussian Splatting (3DGS) is becoming a practical representation for novel view synthesis, but its growing adoption, together with rapid advances in in

- **GuardMarkGS** [arXiv:2605.12919](https://arxiv.org/abs/2605.12919) (arXiv 2026) — 3D Gaussian Splatting (3DGS) is becoming a practical representation for novel view synthesis, but its growing adoption, together with rapid advances in in

- **NGS-Marker** [arXiv:2608.17447](https://arxiv.org/abs/2608.17447) (arXiv 2026) — Novel native watermarking framework for 3DGS

- **NGS-Marker** [arXiv:2608.17447](https://arxiv.org/abs/2608.17447) (arXiv 2026) — Novel native watermarking framework for 3DGS

- **PrinGS** [arXiv:2610.36168](https://arxiv.org/abs/2610.36168) (arXiv 2026) — Representation-level anonymization that directly anonymizes trained splats from a user reference view, removing sensitive content from all novel views in under

- **PrinGS** [arXiv:2610.36168](https://arxiv.org/abs/2610.36168) (arXiv 2026) — Representation-level anonymization that directly anonymizes trained splats from a user reference view, removing sensitive content from all novel views in under

- **3DGEER** [no arXiv ID; verify primary source] (ICLR 2026) — Exact ray-Gaussian integration replacing splatting approximation for fisheye/equirectangular cameras with closed-form 3D mean integral - **3DGS** [arXiv:2308.04079](https://arx

- **3DGEER** [no arXiv ID; verify primary source] (ICLR 2026) — Exact ray-Gaussian integration replacing splatting approximation for fisheye/equirectangular cameras with closed-form 3D mean integral - **3DGS** [arXiv:2308.04079](https://arx

- **3DGS** [arXiv:2308.04079](https://arxiv.org/abs/2308.04079) (SIGGRAPH 2023) — Anisotropic 3D Gaussians with tile-based differentiable rasterization

- **3DGS** [arXiv:2308.04079](https://arxiv.org/abs/2308.04079) (SIGGRAPH 2023) — Anisotropic 3D Gaussians with tile-based differentiable rasterization

- **3DSGS** [arXiv:2605.18334](https://arxiv.org/abs/2605.18334) (arXiv 2026) — General 3D Skew Gaussian primitives with re-derived CUDA rasterization for asymmetric shape modeling

- **3DSGS** [arXiv:2605.18334](https://arxiv.org/abs/2605.18334) (arXiv 2026) — General 3D Skew Gaussian primitives with re-derived CUDA rasterization for asymmetric shape modeling

- **6DGS** [arXiv:2410.04974](https://arxiv.org/abs/2410.04974) (ECCV 2024) — 6-DoF Gaussian Splatting: explicit orientation-aware primitive with full 6D pose parameterization

- **6DGS** [arXiv:2410.04974](https://arxiv.org/abs/2410.04974) (ECCV 2024) — 6-DoF Gaussian Splatting: explicit orientation-aware primitive with full 6D pose parameterization

- **Argus** [no arXiv ID; verify primary source] (arXiv) — Image-derived LiDAR-level pose constraints for 3DGS without LiDAR; consumer-grade 3DGS applications (如视) - **Cookbook of 3D Vision** [arXiv:2606.04291](https://arxiv.org/abs/2606.04

- **Argus** [no arXiv ID; verify primary source] (arXiv) — Image-derived LiDAR-level pose constraints for 3DGS without LiDAR; consumer-grade 3DGS applications (如视) - **Cookbook of 3D Vision** [arXiv:2606.04291](https://arxiv.org/abs/2606.04

- **Cookbook of 3D Vision** [arXiv:2606.04291](https://arxiv.org/abs/2606.04291) (arXiv 2026) — Data-centric taxonomy of 3D vision: point clouds, meshes, voxels, 3D Gaussians; dataset design

- **Cookbook of 3D Vision** [arXiv:2606.04291](https://arxiv.org/abs/2606.04291) (arXiv 2026) — Data-centric taxonomy of 3D vision: point clouds, meshes, voxels, 3D Gaussians; dataset design

- **DisC-GS** [arXiv:2405.15196](https://arxiv.org/abs/2405.15196) (NeurIPS 2024) — Progressive low-pass + discontinuity boundary detection preventing splat artifacts at edges

- **DisC-GS** [arXiv:2405.15196](https://arxiv.org/abs/2405.15196) (NeurIPS 2024) — Progressive low-pass + discontinuity boundary detection preventing splat artifacts at edges

- **DP-GES** [arXiv:2605.25345](https://arxiv.org/abs/2605.25345) (arXiv) — Depth peeling for high-fidelity Gaussian-enhanced surfel rendering; resolves transparency ordering

- **DP-GES** [arXiv:2605.25345](https://arxiv.org/abs/2605.25345) (arXiv) — Depth peeling for high-fidelity Gaussian-enhanced surfel rendering; resolves transparency ordering

- **E3DGS** [arXiv:2607.15536](https://arxiv.org/abs/2607.15536) (arXiv 2026) — Unified geometric-photometric equivariance for 3DGS via Color-as-Geometry embedding; builds SE(3)-equivariant architectures on Gaussian primitives by treating c

- **E3DGS** [arXiv:2607.15536](https://arxiv.org/abs/2607.15536) (arXiv 2026) — Unified geometric-photometric equivariance for 3DGS via Color-as-Geometry embedding; builds SE(3)-equivariant architectures on Gaussian primitives by treating c

- **Fast 3D Foundation Model Initialized GS** [no arXiv ID; verify primary source] (arXiv) — 3D foundation model replacing SfM for pose/point cloud initialization; joint pose-Gaussian optimization; ~3 min/scene training - **GaussianImage** 

- **Fast 3D Foundation Model Initialized GS** [no arXiv ID; verify primary source] (arXiv) — 3D foundation model replacing SfM for pose/point cloud initialization; joint pose-Gaussian optimization; ~3 min/scene training - **GaussianImage** 

- **GaussianImage** [arXiv:2403.08551](https://arxiv.org/abs/2403.08551) (ECCV 2024) — 2D Gaussian image codec at 1000+ FPS

- **GaussianImage** [arXiv:2403.08551](https://arxiv.org/abs/2403.08551) (ECCV 2024) — 2D Gaussian image codec at 1000+ FPS

- **GaussRender** [arXiv:2502.05040](https://arxiv.org/abs/2502.05040) (CVPR 2025) — Unified rendering pipeline for GS supporting multi-modal output (RGB/D/N/S)

- **GaussRender** [arXiv:2502.05040](https://arxiv.org/abs/2502.05040) (CVPR 2025) — Unified rendering pipeline for GS supporting multi-modal output (RGB/D/N/S)

- **MetaPoint** [arXiv:2606.05031](https://arxiv.org/abs/2606.05031) (arXiv) — Precise spatial control in agentic visual generation; represents continuous 2D coordinate as special token

- **MetaPoint** [arXiv:2606.05031](https://arxiv.org/abs/2606.05031) (arXiv) — Precise spatial control in agentic visual generation; represents continuous 2D coordinate as special token

- **Mip-Splatting** [arXiv:2311.16493](https://arxiv.org/abs/2311.16493) (CVPR 2024 (Best Student Paper)) — Anti-aliased 3DGS with 3D smoothing + 2D Mip filter

- **Mip-Splatting** [arXiv:2311.16493](https://arxiv.org/abs/2311.16493) (CVPR 2024 (Best Student Paper)) — Anti-aliased 3DGS with 3D smoothing + 2D Mip filter

- **Normal-GS** [arXiv:2410.20593](https://arxiv.org/abs/2410.20593) (NeurIPS 2024) — Normal-involved rendering: normal constraint + differentiable normal guiding splat distribution

- **Normal-GS** [arXiv:2410.20593](https://arxiv.org/abs/2410.20593) (NeurIPS 2024) — Normal-involved rendering: normal constraint + differentiable normal guiding splat distribution

- **ODGS** [arXiv:2410.20686](https://arxiv.org/abs/2410.20686) (NeurIPS 2024) — Spherical projection + panoramic camera GS rasterization adaptation for 360-degree images

- **ODGS** [arXiv:2410.20686](https://arxiv.org/abs/2410.20686) (NeurIPS 2024) — Spherical projection + panoramic camera GS rasterization adaptation for 360-degree images

- **PRISM3D** [no arXiv ID; verify primary source] (arXiv) — First framework bootstrapping 3DGS from extreme motion blur; MCMC probabilistic densification + continuous Bezier trajectory physics modeling - **PureLight** [arXiv:2606.04319](ht

- **PRISM3D** [no arXiv ID; verify primary source] (arXiv) — First framework bootstrapping 3DGS from extreme motion blur; MCMC probabilistic densification + continuous Bezier trajectory physics modeling - **PureLight** [arXiv:2606.04319](ht

- **PureLight** [arXiv:2606.04319](https://arxiv.org/abs/2606.04319) (arXiv) — Learning complex luminaires with light tracing; forward light tracing for efficient difficult light paths

- **PureLight** [arXiv:2606.04319](https://arxiv.org/abs/2606.04319) (arXiv) — Learning complex luminaires with light tracing; forward light tracing for efficient difficult light paths

- **SalientGS** [arXiv:2607.11285](https://arxiv.org/abs/2607.11285) (arXiv) — Unified SfM-to-3DGS pipeline with importance-guided MCMC Gaussian allocation; reallocates capacity from well-fit to underfit regions; 15-min end-to-end SOTA

- **SalientGS** [arXiv:2607.11285](https://arxiv.org/abs/2607.11285) (arXiv) — Unified SfM-to-3DGS pipeline with importance-guided MCMC Gaussian allocation; reallocates capacity from well-fit to underfit regions; 15-min end-to-end SOTA

- **SAM 3D** [no arXiv ID; verify primary source] (CVPR 2026 Honorable Mention) — Foundation model for single-image 3D reconstruction; generalizable 3D prior enables diverse downstream tasks - **SNS** [arXiv:2605.15010](https://arxiv.org/ab

- **SAM 3D** [no arXiv ID; verify primary source] (CVPR 2026 Honorable Mention) — Foundation model for single-image 3D reconstruction; generalizable 3D prior enables diverse downstream tasks - **SNS** [arXiv:2605.15010](https://arxiv.org/ab

- **SNS** [arXiv:2605.15010](https://arxiv.org/abs/2605.15010) (arXiv) — 3D Gaussian Splatting (3DGS) has emerged as a leading representation for real-time novel view synthesis and has been widely adopted in various downstream applications.

- **SNS** [arXiv:2605.15010](https://arxiv.org/abs/2605.15010) (arXiv) — 3D Gaussian Splatting (3DGS) has emerged as a leading representation for real-time novel view synthesis and has been widely adopted in various downstream applications.

- **Softmax-GS** [arXiv:2604.27437](https://arxiv.org/abs/2604.27437) (CVPR 2026 Findings) — Learnable softmax-based competition replacing α-compositing

- **Softmax-GS** [arXiv:2604.27437](https://arxiv.org/abs/2604.27437) (CVPR 2026 Findings) — Learnable softmax-based competition replacing α-compositing

- **SpatialQ** [arXiv:2607.26595](https://arxiv.org/abs/2607.26595) (arXiv 2026) — 3D Gaussian Splatting (3DGS) has emerged as an effective representation for novel view synthesis and 3D scene reconstruction, creating an increasing demand f

- **SpatialQ** [arXiv:2607.26595](https://arxiv.org/abs/2607.26595) (arXiv 2026) — 3D Gaussian Splatting (3DGS) has emerged as an effective representation for novel view synthesis and 3D scene reconstruction, creating an increasing demand f

- **Topo-GS** [arXiv:2605.17011](https://arxiv.org/abs/2605.17011) (arXiv 2026) — Topological GS for high-dimensional data embedding; ARAP prior + tangent space alignment via Procrustes; topology-aware loss

- **Topo-GS** [arXiv:2605.17011](https://arxiv.org/abs/2605.17011) (arXiv 2026) — Topological GS for high-dimensional data embedding; ARAP prior + tangent space alignment via Procrustes; topology-aware loss

- **UniGS** [arXiv:2510.12174](https://arxiv.org/abs/2510.12174) (CVPR 2025) — Unified Gaussian Splatting: single model supporting multiple rendering modes (RGB/depth/semantic)

- **UniGS** [arXiv:2510.12174](https://arxiv.org/abs/2510.12174) (CVPR 2025) — Unified Gaussian Splatting: single model supporting multiple rendering modes (RGB/depth/semantic)

- **UniTriSplat** [arXiv:2606.29794](https://arxiv.org/abs/2606.29794) (arXiv 2026) — To address this limitation, we propose UniTriSplat, a unified 3DGS framework for universal cameras that reformulates Gaussian splatting on the unit sphere

- **UniTriSplat** [arXiv:2606.29794](https://arxiv.org/abs/2606.29794) (arXiv 2026) — To address this limitation, we propose UniTriSplat, a unified 3DGS framework for universal cameras that reformulates Gaussian splatting on the unit sphere

- **3D-GIMP** [arXiv:2607.20789](https://arxiv.org/abs/2607.20789) (arXiv 2026) — 3DGS inpainting + PatchMatch hybrid; single generative inpainting on key view then propagate

- **3D-GIMP** [arXiv:2607.20789](https://arxiv.org/abs/2607.20789) (arXiv 2026) — 3DGS inpainting + PatchMatch hybrid; single generative inpainting on key view then propagate

- **AOV-GS** [arXiv:2610.29374](https://arxiv.org/abs/2610.29374) (arXiv 2026) — Active 3D Gaussian mapping system guiding exploration with frozen foundation models; CLIP mask-embedding novelty re-ranks candidate views toward semantically i

- **AOV-GS** [arXiv:2610.29374](https://arxiv.org/abs/2610.29374) (arXiv 2026) — Active 3D Gaussian mapping system guiding exploration with frozen foundation models; CLIP mask-embedding novelty re-ranks candidate views toward semantically i

- **CapFrame** [arXiv:2608.30342](https://arxiv.org/abs/2608.30342) (ECCV 2026) — Text-instructed viewpoint localization in 3D Gaussian scenes; geometric pseudo-labels bridge language queries and spatial grounding

- **CapFrame** [arXiv:2608.30342](https://arxiv.org/abs/2608.30342) (ECCV 2026) — Text-instructed viewpoint localization in 3D Gaussian scenes; geometric pseudo-labels bridge language queries and spatial grounding

- **CausalSplat** [arXiv:2608.11150](https://arxiv.org/abs/2608.11150) (arXiv) — Hierarchical causal reasoning in 3DGS; implicit intent understanding + complex spatial relationship comprehension for scene-level causal inference

- **CausalSplat** [arXiv:2608.11150](https://arxiv.org/abs/2608.11150) (arXiv) — Hierarchical causal reasoning in 3DGS; implicit intent understanding + complex spatial relationship comprehension for scene-level causal inference

- **CDSeg** [arXiv:2608.05482](https://arxiv.org/abs/2608.05482) (arXiv 2026) — Cross-Domain Segmentation via Gaussian Splatting (CDSeg), a label-transfer interface that requires no task-specific 3D segmentation training and uses Gaussian p

- **CDSeg** [arXiv:2608.05482](https://arxiv.org/abs/2608.05482) (arXiv 2026) — Cross-Domain Segmentation via Gaussian Splatting (CDSeg), a label-transfer interface that requires no task-specific 3D segmentation training and uses Gaussian p

- **ClunoGS** [arXiv:2610.30013](https://arxiv.org/abs/2610.30013) (arXiv 2026) — Compact framework for Language 3DGS that combines semantics-aware Gaussian sparsification and vector quantization in a unified constrained objective

- **ClunoGS** [arXiv:2610.30013](https://arxiv.org/abs/2610.30013) (arXiv 2026) — Compact framework for Language 3DGS that combines semantics-aware Gaussian sparsification and vector quantization in a unified constrained objective

- **ClusterSplat** [arXiv:2610.36462](https://arxiv.org/abs/2610.36462) (arXiv 2026) — Reframes 3D target selection in GS grounding as cluster-level selection over scene-adaptive candidates; instance-aware Gaussian features and description-

- **ClusterSplat** [arXiv:2610.36462](https://arxiv.org/abs/2610.36462) (arXiv 2026) — Reframes 3D target selection in GS grounding as cluster-level selection over scene-adaptive candidates; instance-aware Gaussian features and description-

- **Consistent Scene Understanding in 3DGS** [no arXiv ID; verify primary source] (arXiv) — Multi-cue mask refinement with semantic/depth/edge fusion + cross-view mask matching for consistent 3DGS instance segmentation - **CoRef-GS** [arXiv

- **Consistent Scene Understanding in 3DGS** [no arXiv ID; verify primary source] (arXiv) — Multi-cue mask refinement with semantic/depth/edge fusion + cross-view mask matching for consistent 3DGS instance segmentation - **CoRef-GS** [arXiv

- **CoRef-GS** [arXiv:2609.20586](https://arxiv.org/abs/2609.20586) (arXiv 2026) — Cooperative referring Gaussian splatting framework

- **CoRef-GS** [arXiv:2609.20586](https://arxiv.org/abs/2609.20586) (arXiv 2026) — Cooperative referring Gaussian splatting framework

- **DGSG-Mind** [arXiv:2605.29879](https://arxiv.org/abs/2605.29879) (arXiv) — Dynamic 3D Gaussian scene graphs with probabilistic voxel grid + 3D Gaussian Mind for embodied reasoning

- **DGSG-Mind** [arXiv:2605.29879](https://arxiv.org/abs/2605.29879) (arXiv) — Dynamic 3D Gaussian scene graphs with probabilistic voxel grid + 3D Gaussian Mind for embodied reasoning

- **DIGS** [arXiv:2610.16177](https://arxiv.org/abs/2610.16177) (arXiv 2026) — Distribution-informed Gaussian splatting for training-free open-vocabulary 3D segmentation; boundary Gaussians carry feature distributions instead of single feat

- **DIGS** [arXiv:2610.16177](https://arxiv.org/abs/2610.16177) (arXiv 2026) — Distribution-informed Gaussian splatting for training-free open-vocabulary 3D segmentation; boundary Gaussians carry feature distributions instead of single feat

- **Feature 3DGS** [arXiv:2312.03203](https://arxiv.org/abs/2312.03203) (CVPR 2024) — Distilled DINO/SAM features for 3D segmentation/detection

- **Feature 3DGS** [arXiv:2312.03203](https://arxiv.org/abs/2312.03203) (CVPR 2024) — Distilled DINO/SAM features for 3D segmentation/detection

- **Gaga** [no arXiv ID; verify primary source] (arXiv) — Group Any Gaussians via 3D-aware Memory Bank; leverages inconsistent 2D masks from zero-shot segmentation models; 3D-aware memory bank for robust cross-pose mask association - **Gaus

- **Gaga** [no arXiv ID; verify primary source] (arXiv) — Group Any Gaussians via 3D-aware Memory Bank; leverages inconsistent 2D masks from zero-shot segmentation models; 3D-aware memory bank for robust cross-pose mask association - **Gaus

- **GaussianGraph** [arXiv:2503.04034](https://arxiv.org/abs/2503.04034) (arXiv) — 3DGS-based scene graph generation with Control-Follow clustering + 3D spatial relation correction for open-world understanding

- **GaussianGraph** [arXiv:2503.04034](https://arxiv.org/abs/2503.04034) (arXiv) — 3DGS-based scene graph generation with Control-Follow clustering + 3D spatial relation correction for open-world understanding

- **GeoSAM-3D** [arXiv:2606.00447](https://arxiv.org/abs/2606.00447) (arXiv 2026) — Open-vocabulary 3D scene segmentation on monocular video: 3DGS reconstruction + differentiable graph-geodesic propagation kernel (SAM prompt propagates alon

- **GeoSAM-3D** [arXiv:2606.00447](https://arxiv.org/abs/2606.00447) (arXiv 2026) — Open-vocabulary 3D scene segmentation on monocular video: 3DGS reconstruction + differentiable graph-geodesic propagation kernel (SAM prompt propagates alon

- **GLMap** [arXiv:2605.01736](https://arxiv.org/abs/2605.01736) (CVPR 2026) — Multi-scale Gaussian-Language map for zero-shot navigation

- **GLMap** [arXiv:2605.01736](https://arxiv.org/abs/2605.01736) (CVPR 2026) — Multi-scale Gaussian-Language map for zero-shot navigation

- **GroupForward** [arXiv:2608.17535](https://arxiv.org/abs/2608.17535) (arXiv 2026) — To this end, we propose GroupForward, an instance-grouped feed-forward Gaussian splatting model that reconstructs geometry, appearance, instance structur

- **GroupForward** [arXiv:2608.17535](https://arxiv.org/abs/2608.17535) (arXiv 2026) — To this end, we propose GroupForward, an instance-grouped feed-forward Gaussian splatting model that reconstructs geometry, appearance, instance structur

- **GS-LLM** [arXiv:2412.09176](https://arxiv.org/abs/2412.09176) (CVPR 2025) — LLM-guided GS for reasoning-driven 3D scene understanding and manipulation

- **GS-LLM** [arXiv:2412.09176](https://arxiv.org/abs/2412.09176) (CVPR 2025) — LLM-guided GS for reasoning-driven 3D scene understanding and manipulation

- **HumanSkill-4D** [arXiv:2610.24452](https://arxiv.org/abs/2610.24452) (arXiv preprint) — Abstract context: We introduce avatar referring segmentation, which selects regions of animatable Gaussian avatars from language queries.

- **HumanSkill-4D** [arXiv:2610.24452](https://arxiv.org/abs/2610.24452) (arXiv preprint) — Abstract context: We introduce avatar referring segmentation, which selects regions of animatable Gaussian avatars from language queries.

- **Ilov3Splat** [arXiv:2605.04506](https://arxiv.org/abs/2605.04506) (ICPR 2026) — Instance-level open-vocabulary 3DGS via multi-resolution hash embedding for CLIP features + SAM contrastive instance field; two-stage 3D clustering for natu

- **Ilov3Splat** [arXiv:2605.04506](https://arxiv.org/abs/2605.04506) (ICPR 2026) — Instance-level open-vocabulary 3DGS via multi-resolution hash embedding for CLIP features + SAM contrastive instance field; two-stage 3D clustering for natu

- **LangSplat** [arXiv:2312.16084](https://arxiv.org/abs/2312.16084) (CVPR 2024) — CLIP features stored per-Gaussian for open-vocabulary 3D queries

- **LangSplat** [arXiv:2312.16084](https://arxiv.org/abs/2312.16084) (CVPR 2024) — CLIP features stored per-Gaussian for open-vocabulary 3D queries

- **LB-Edit** [arXiv:2607.19777](https://arxiv.org/abs/2607.19777) (arXiv 2026) — Attention-guided editing camera placement (ACP) + multi-view attention alignment (MAA); 5 views, 7x lower latency

- **LB-Edit** [arXiv:2607.19777](https://arxiv.org/abs/2607.19777) (arXiv 2026) — Attention-guided editing camera placement (ACP) + multi-view attention alignment (MAA); 5 views, 7x lower latency

- **LEGO** [arXiv:2608.10057](https://arxiv.org/abs/2608.10057) (ECCV 2026) — Hierarchical language gaussian splatting; multi-view SAM granularity-adaptive re-grading into 3D-consistent hierarchy; per-level language scene graph supports LLM

- **LEGO** [arXiv:2608.10057](https://arxiv.org/abs/2608.10057) (ECCV 2026) — Hierarchical language gaussian splatting; multi-view SAM granularity-adaptive re-grading into 3D-consistent hierarchy; per-level language scene graph supports LLM

- **LegoOcc++** [arXiv:2610.34093](https://arxiv.org/abs/2610.34093) (arXiv 2026) — Open-vocabulary indoor 3D occupancy prediction built on Language-Embedded Gaussians; opacity-aware Poisson Gaussian-to-Occupancy formulation enables volumet

- **LegoOcc++** [arXiv:2610.34093](https://arxiv.org/abs/2610.34093) (arXiv 2026) — Open-vocabulary indoor 3D occupancy prediction built on Language-Embedded Gaussians; opacity-aware Poisson Gaussian-to-Occupancy formulation enables volumet

- **NG-GS** [arXiv:2604.14706](https://arxiv.org/abs/2604.14706) (arXiv) — NeRF-guided Gaussian-level semantic assignment

- **NG-GS** [arXiv:2604.14706](https://arxiv.org/abs/2604.14706) (arXiv) — NeRF-guided Gaussian-level semantic assignment

- **NormLift** [arXiv:2609.18898](https://arxiv.org/abs/2609.18898) (arXiv 2026) — 3D-side feature lifting formulating per-Gaussian assignment as cosine alignment for semantically reliable Gaussians.

- **NormLift** [arXiv:2609.18898](https://arxiv.org/abs/2609.18898) (arXiv 2026) — 3D-side feature lifting formulating per-Gaussian assignment as cosine alignment for semantically reliable Gaussians.

- **OP2GS** [arXiv:2605.20044](https://arxiv.org/abs/2605.20044) (arXiv 2026) — Dual-opacity primitives decoupling visual existence from instance occupancy for object-aware 3DGS

- **OP2GS** [arXiv:2605.20044](https://arxiv.org/abs/2605.20044) (arXiv 2026) — Dual-opacity primitives decoupling visual existence from instance occupancy for object-aware 3DGS

- **OpenGaFF** [arXiv:2605.06088](https://arxiv.org/abs/2605.06088) (arXiv) — Understanding open-vocabulary 3D scenes with Gaussian-based representations remains challenging due to fragmented and spatially inconsistent semantic predictions 

- **OpenGaFF** [arXiv:2605.06088](https://arxiv.org/abs/2605.06088) (arXiv) — Understanding open-vocabulary 3D scenes with Gaussian-based representations remains challenging due to fragmented and spatially inconsistent semantic predictions 

- **OpenGaussian** [arXiv:2406.02058](https://arxiv.org/abs/2406.02058) (NeurIPS 2024) — Per-Gaussian feature distillation for point-level open-vocabulary 3D understanding

- **OpenGaussian** [arXiv:2406.02058](https://arxiv.org/abs/2406.02058) (NeurIPS 2024) — Per-Gaussian feature distillation for point-level open-vocabulary 3D understanding

- **OpenSplatGraph** [arXiv:2610.07569](https://arxiv.org/abs/2610.07569) (arXiv 2026) — Unified framework that constructs persistent 3D scene graphs directly from an online Gaussian-based open-vocabulary semantic map, augmenting it with a 

- **OpenSplatGraph** [arXiv:2610.07569](https://arxiv.org/abs/2610.07569) (arXiv 2026) — Unified framework that constructs persistent 3D scene graphs directly from an online Gaussian-based open-vocabulary semantic map, augmenting it with a 

- **OutLangSplat** [arXiv:2608.04560](https://arxiv.org/abs/2608.04560) (arXiv) — UAV outdoor 3D language GS; 2D-3D dual-branch representation; training-free contribution and consistency-aware gaussian feature aggregation; first UAV open-vo

- **OutLangSplat** [arXiv:2608.04560](https://arxiv.org/abs/2608.04560) (arXiv) — UAV outdoor 3D language GS; 2D-3D dual-branch representation; training-free contribution and consistency-aware gaussian feature aggregation; first UAV open-vo

- **OV-Foam** [arXiv:2610.28131](https://arxiv.org/abs/2610.28131) (arXiv preprint) — Abstract context: Existing feature lifting approaches for open-vocabulary segmentation rely on simple backprojection to assign features to primitives.

- **OV-Foam** [arXiv:2610.28131](https://arxiv.org/abs/2610.28131) (arXiv preprint) — Abstract context: Existing feature lifting approaches for open-vocabulary segmentation rely on simple backprojection to assign features to primitives.

- **PointGS** [arXiv:2605.11520](https://arxiv.org/abs/2605.11520) (CVPR 2026) — Unsupervised point cloud segmentation is critical for embodied artificial intelligence and autonomous driving, as it mitigates the prohibitive cost of dense po

- **PointGS** [arXiv:2605.11520](https://arxiv.org/abs/2605.11520) (CVPR 2026) — Unsupervised point cloud segmentation is critical for embodied artificial intelligence and autonomous driving, as it mitigates the prohibitive cost of dense po

- **RecallGS** [arXiv:2610.31293](https://arxiv.org/abs/2610.31293) (arXiv 2026) — Lightweight training-free recall mechanism for online semantic GS; treats past observations as a non-parametric geometric-semantic prior via a recall-ground-

- **RecallGS** [arXiv:2610.31293](https://arxiv.org/abs/2610.31293) (arXiv 2026) — Lightweight training-free recall mechanism for online semantic GS; treats past observations as a non-parametric geometric-semantic prior via a recall-ground-

- **ReferGaussian** [arXiv:2610.30326](https://arxiv.org/abs/2610.30326) (arXiv 2026) — Unified framework built upon 4DGS for referring segmentation in dynamic scenes; converts low-level dynamic Gaussians into persistent language-addressabl

- **ReferGaussian** [arXiv:2610.30326](https://arxiv.org/abs/2610.30326) (arXiv 2026) — Unified framework built upon 4DGS for referring segmentation in dynamic scenes; converts low-level dynamic Gaussians into persistent language-addressabl

- **ReferSplat** [no arXiv ID; verify primary source] (ICML 2025 Oral) — Referring segmentation in 3D Gaussian Splatting with language-guided instance-level features for open-vocabulary 3D understanding - **ReLaGS** [arXiv:2603.17605](https

- **ReferSplat** [no arXiv ID; verify primary source] (ICML 2025 Oral) — Referring segmentation in 3D Gaussian Splatting with language-guided instance-level features for open-vocabulary 3D understanding - **ReLaGS** [arXiv:2603.17605](https

- **ReLaGS** [arXiv:2603.17605](https://arxiv.org/abs/2603.17605) (CVPR 2026) — Relational Language Gaussian Splatting for open-vocabulary 3D scene understanding

- **ReLaGS** [arXiv:2603.17605](https://arxiv.org/abs/2603.17605) (CVPR 2026) — Relational Language Gaussian Splatting for open-vocabulary 3D scene understanding

- **RelateGS** [arXiv:2610.33868](https://arxiv.org/abs/2610.33868) (arXiv 2026) — Structured language-aware Gaussian representation selectively refining semantic and relational evidence for 3D visual grounding; suppresses unreliable observ

- **RelateGS** [arXiv:2610.33868](https://arxiv.org/abs/2610.33868) (arXiv 2026) — Structured language-aware Gaussian representation selectively refining semantic and relational evidence for 3D visual grounding; suppresses unreliable observ

- **Robust Prior-Guided Seg.** [arXiv:2605.16065](https://arxiv.org/abs/2605.16065) (ICIP) — SAM-HQ + prior-guided label reassignment for multiview-consistent 3DGS segmentation; object removal/extraction/recoloring

- **Robust Prior-Guided Seg.** [arXiv:2605.16065](https://arxiv.org/abs/2605.16065) (ICIP) — SAM-HQ + prior-guided label reassignment for multiview-consistent 3DGS segmentation; object removal/extraction/recoloring

- **SAGO** [no arXiv ID; verify primary source] (arXiv) — Online segment 3D Gaussians via virtual drones; reframes interactive 3DGS segmentation as Next-Best-View planning; 50x faster than prior methods - **SceneSplat** [no arXiv ID; verify

- **SAGO** [no arXiv ID; verify primary source] (arXiv) — Online segment 3D Gaussians via virtual drones; reframes interactive 3DGS segmentation as Next-Best-View planning; 50x faster than prior methods - **SceneSplat** [no arXiv ID; verify

- **SceneSplat** [no arXiv ID; verify primary source] (ICCV 2025 Oral) — Gaussian Splatting-based scene understanding with vision-language pretraining; SceneSplat-7K dataset; native 3DGS encoder for open-vocabulary segmentation - **SCOUP** 

- **SceneSplat** [no arXiv ID; verify primary source] (ICCV 2025 Oral) — Gaussian Splatting-based scene understanding with vision-language pretraining; SceneSplat-7K dataset; native 3DGS encoder for open-vocabulary segmentation - **SCOUP** 

- **SCOUP** [arXiv:2605.13600](https://arxiv.org/abs/2605.13600) (arXiv 2026) — Sparse Code Uplifting for efficient 3D language GS with 400x training speedup

- **SCOUP** [arXiv:2605.13600](https://arxiv.org/abs/2605.13600) (arXiv 2026) — Sparse Code Uplifting for efficient 3D language GS with 400x training speedup

- **Semantic Foam** [arXiv:2604.26262](https://arxiv.org/abs/2604.26262) (CVPR 2026 Highlight) — Volumetric Voronoi mesh for spatial + semantic decomposition

- **Semantic Foam** [arXiv:2604.26262](https://arxiv.org/abs/2604.26262) (CVPR 2026 Highlight) — Volumetric Voronoi mesh for spatial + semantic decomposition

- **TrackRef3D** [arXiv:2605.26576](https://arxiv.org/abs/2605.26576) (arXiv) — Track-then-label paradigm with TSCM for open-world referring segmentation in 3DGS

- **TrackRef3D** [arXiv:2605.26576](https://arxiv.org/abs/2605.26576) (arXiv) — Track-then-label paradigm with TSCM for open-world referring segmentation in 3DGS

- **Training-Free Target-Aware Refinement of 3D Gaussian Language Map** [arXiv:2610.21619](https://arxiv.org/abs/2610.21619) (arXiv preprint) — Abstract context: Distilling vision-language features into 3D Gaussian Splatting yields a langua

- **Training-Free Target-Aware Refinement of 3D Gaussian Language Map** [arXiv:2610.21619](https://arxiv.org/abs/2610.21619) (arXiv preprint) — Abstract context: Distilling vision-language features into 3D Gaussian Splatting yields a langua

- **VCAR** [arXiv:2608.30870](https://arxiv.org/abs/2608.30870) (ACM MM 2026) — Training-free 3DGS segmentation via view completeness and axis-aware boundary refinement; no additional training required for high-quality 3D segmentation

- **VCAR** [arXiv:2608.30870](https://arxiv.org/abs/2608.30870) (ACM MM 2026) — Training-free 3DGS segmentation via view completeness and axis-aware boundary refinement; no additional training required for high-quality 3D segmentation

- **VLRC** [no arXiv ID; verify primary source] (arXiv) — Vision-Language Reprojection Consistency as scalable signal for better feed-forward 3D pretraining - **X-GS** [arXiv:2603.09632](https://arxiv.org/abs/2603.09632) (arXiv) — Extensibl

- **VLRC** [no arXiv ID; verify primary source] (arXiv) — Vision-Language Reprojection Consistency as scalable signal for better feed-forward 3D pretraining - **X-GS** [arXiv:2603.09632](https://arxiv.org/abs/2603.09632) (arXiv) — Extensibl

- **X-GS** [arXiv:2603.09632](https://arxiv.org/abs/2603.09632) (arXiv) — Extensible GS framework: X-GS-Perceiver for SLAM + semantic distillation for downstream tasks

- **X-GS** [arXiv:2603.09632](https://arxiv.org/abs/2603.09632) (arXiv) — Extensible GS framework: X-GS-Perceiver for SLAM + semantic distillation for downstream tasks

- **ZeroSplat** [arXiv:2607.18801](https://arxiv.org/abs/2607.18801) (ECCV 2026) — Generalized referring 3DGS segmentation (GR3DGS); zero-feature, training-free, 0/1/N target support

- **ZeroSplat** [arXiv:2607.18801](https://arxiv.org/abs/2607.18801) (ECCV 2026) — Generalized referring 3DGS segmentation (GR3DGS); zero-feature, training-free, 0/1/N target support

- **3DGS³** [arXiv:2605.11489](https://arxiv.org/abs/2605.11489) (arXiv) — Joint super sampling and frame interpolation for real-time large-scale 3DGS rendering; unified SS+FI pipeline enabling high-fidelity frame generation at scale

- **3DGS³** [arXiv:2605.11489](https://arxiv.org/abs/2605.11489) (arXiv) — Joint super sampling and frame interpolation for real-time large-scale 3DGS rendering; unified SS+FI pipeline enabling high-fidelity frame generation at scale

- **3DGStream** [arXiv:2403.01444](https://arxiv.org/abs/2403.01444) (CVPR 2024) — 3D Gaussian Stream for real-time radiance field streaming; per-frame 3DGS optimization with temporal consistency for long dynamic video sequences; per-frame 

- **3DGStream** [arXiv:2403.01444](https://arxiv.org/abs/2403.01444) (CVPR 2024) — 3D Gaussian Stream for real-time radiance field streaming; per-frame 3DGS optimization with temporal consistency for long dynamic video sequences; per-frame 

- **4DGen** [arXiv:2312.17225](https://arxiv.org/abs/2312.17225) (arXiv preprint) — 4D content creation as monocular static-to-dynamic lifting with reference-anchored score distillation + Generative Gaussian Prior initialized from depth-poi

- **4DGen** [arXiv:2312.17225](https://arxiv.org/abs/2312.17225) (arXiv preprint) — 4D content creation as monocular static-to-dynamic lifting with reference-anchored score distillation + Generative Gaussian Prior initialized from depth-poi

- **4DGS** [arXiv:2310.08528](https://arxiv.org/abs/2310.08528) (CVPR 2024) — 4D anisotropic Gaussians (3D + time) with regularized deformation

- **4DGS** [arXiv:2310.08528](https://arxiv.org/abs/2310.08528) (CVPR 2024) — 4D anisotropic Gaussians (3D + time) with regularized deformation

- **4DGS-Fixer** [arXiv:2609.21176](https://arxiv.org/abs/2609.21176) (arXiv 2026) — Generative sparse-view 4D Gaussian Splatting with iterative refinement guided by video diffusion priors.

- **4DGS-Fixer** [arXiv:2609.21176](https://arxiv.org/abs/2609.21176) (arXiv 2026) — Generative sparse-view 4D Gaussian Splatting with iterative refinement guided by video diffusion priors.

- **4D-rotor GS** [arXiv:2402.03307](https://arxiv.org/abs/2402.03307) (SIGGRAPH 2024) — 4D rotor-based Gaussian Splatting; geometric algebra rotors for compact 4D motion representation suitable for articulated and non-rigid dynamics; evalu

- **4D-rotor GS** [arXiv:2402.03307](https://arxiv.org/abs/2402.03307) (SIGGRAPH 2024) — 4D rotor-based Gaussian Splatting; geometric algebra rotors for compact 4D motion representation suitable for articulated and non-rigid dynamics; evalu

- **ACA-GS** [arXiv:2608.04581](https://arxiv.org/abs/2608.04581) (arXiv 2026) — Recent advances in 4D Gaussian Splatting (4DGS) enable high-fidelity, real-time spatiotemporal rendering, but expose a fundamental trade-off between motion exp

- **ACA-GS** [arXiv:2608.04581](https://arxiv.org/abs/2608.04581) (arXiv 2026) — Recent advances in 4D Gaussian Splatting (4DGS) enable high-fidelity, real-time spatiotemporal rendering, but expose a fundamental trade-off between motion exp

- **Amortized Anchor Refinement** [arXiv:2608.30218](https://arxiv.org/abs/2608.30218) (arXiv) — Deployable continuous-time 4D Gaussian reconstruction with amortized anchor refinement; enables real-time temporal coherent 4D reconstruction f

- **Amortized Anchor Refinement** [arXiv:2608.30218](https://arxiv.org/abs/2608.30218) (arXiv) — Deployable continuous-time 4D Gaussian reconstruction with amortized anchor refinement; enables real-time temporal coherent 4D reconstruction f

- **AniGS** [arXiv:2607.18539](https://arxiv.org/abs/2607.18539) (arXiv 2026) — Scene-level 3DGS animation via diffusion prior; canonical 3DGS + time-conditioned deformation; iterative dataset-model update

- **AniGS** [arXiv:2607.18539](https://arxiv.org/abs/2607.18539) (arXiv 2026) — Scene-level 3DGS animation via diffusion prior; canonical 3DGS + time-conditioned deformation; iterative dataset-model update

- **ATGS** [arXiv:2608.30184](https://arxiv.org/abs/2608.30184) (SIGGRAPH 2026) — Anchored Temporal Gaussian Splatting for long volumetric video representation; temporal anchor points stabilize Gaussian trajectories over extended video sequ

- **ATGS** [arXiv:2608.30184](https://arxiv.org/abs/2608.30184) (SIGGRAPH 2026) — Anchored Temporal Gaussian Splatting for long volumetric video representation; temporal anchor points stabilize Gaussian trajectories over extended video sequ

- **BARD-GS** [arXiv:2503.15835](https://arxiv.org/abs/2503.15835) (CVPR 2025) — Bundle-Adjustable and Rational Deformation Gaussian Splatting; rational motion model with bundle adjustment for complex non-rigid dynamics; rational deformatio

- **BARD-GS** [arXiv:2503.15835](https://arxiv.org/abs/2503.15835) (CVPR 2025) — Bundle-Adjustable and Rational Deformation Gaussian Splatting; rational motion model with bundle adjustment for complex non-rigid dynamics; rational deformatio

- **ChainSplat** [arXiv:2608.28570](https://arxiv.org/abs/2608.28570) (arXiv 2026) — Physics-inspired framework that jointly learns the 3D geometry, appearance, kinematics, and dynamics of DLOs solely from multi-view RGB videos

- **ChainSplat** [arXiv:2608.28570](https://arxiv.org/abs/2608.28570) (arXiv 2026) — Physics-inspired framework that jointly learns the 3D geometry, appearance, kinematics, and dynamics of DLOs solely from multi-view RGB videos

- **ClipGStream** [arXiv:2604.13746](https://arxiv.org/abs/2604.13746) (CVPR 2026) — Clip-stream any-length any-motion multi-view dynamic scene reconstruction

- **ClipGStream** [arXiv:2604.13746](https://arxiv.org/abs/2604.13746) (CVPR 2026) — Clip-stream any-length any-motion multi-view dynamic scene reconstruction

- **CoGS** [arXiv:2312.05664](https://arxiv.org/abs/2312.05664) (CVPR 2024) — Controllable Gaussian Splatting for dynamic scenes; invertible deformation network enables both reconstruction and real-time editing of dynamic content; deformati

- **CoGS** [arXiv:2312.05664](https://arxiv.org/abs/2312.05664) (CVPR 2024) — Controllable Gaussian Splatting for dynamic scenes; invertible deformation network enables both reconstruction and real-time editing of dynamic content; deformati

- **Color-Encoded Illumination** [arXiv:2604.26920](https://arxiv.org/abs/2604.26920) (CVPR 2026 Highlight) — High-speed volumetric reconstruction via color-coded illumination

- **Color-Encoded Illumination** [arXiv:2604.26920](https://arxiv.org/abs/2604.26920) (CVPR 2026 Highlight) — High-speed volumetric reconstruction via color-coded illumination

- **Comp4D** [arXiv:2403.16993](https://arxiv.org/abs/2403.16993) (WACV 2026) — Compositional 4D generation: per-object 4D representation with compositional Gaussian streams, decoupled motion modeling from a text-to-video diffusion prior

- **Comp4D** [arXiv:2403.16993](https://arxiv.org/abs/2403.16993) (WACV 2026) — Compositional 4D generation: per-object 4D representation with compositional Gaussian streams, decoupled motion modeling from a text-to-video diffusion prior

- **CubifyGS** [arXiv:2606.28720](https://arxiv.org/abs/2606.28720) (arXiv 2026) — Object-level mapping framework that shifts dynamic maintenance from passive re-optimization to active asset management

- **CubifyGS** [arXiv:2606.28720](https://arxiv.org/abs/2606.28720) (arXiv 2026) — Object-level mapping framework that shifts dynamic maintenance from passive re-optimization to active asset management

- **D2-4DGS** [arXiv:2608.01588](https://arxiv.org/abs/2608.01588) (arXiv 2026) — To exploit their complementarity, we propose D2-4DGS, a sparse-camera dynamic 4D Gaussian Splatting framework guided by dual-source depth priors

- **D2-4DGS** [arXiv:2608.01588](https://arxiv.org/abs/2608.01588) (arXiv 2026) — To exploit their complementarity, we propose D2-4DGS, a sparse-camera dynamic 4D Gaussian Splatting framework guided by dual-source depth priors

- **D4RT** [arXiv:2512.08924](https://arxiv.org/abs/2512.08924) (CVPR 2026 Best Paper) — Unified query mechanism for 4D reconstruction + tracking from monocular video; 200+ FPS

- **D4RT** [arXiv:2512.08924](https://arxiv.org/abs/2512.08924) (CVPR 2026 Best Paper) — Unified query mechanism for 4D reconstruction + tracking from monocular video; 200+ FPS

- **Deformable-3DGS** [arXiv:2309.13101](https://arxiv.org/abs/2309.13101) (CVPR 2024) — Deformation field network for 3DGS enabling high-fidelity dynamic scene rendering

- **Deformable-3DGS** [arXiv:2309.13101](https://arxiv.org/abs/2309.13101) (CVPR 2024) — Deformation field network for 3DGS enabling high-fidelity dynamic scene rendering

- **DeGenseGS** [no arXiv ID; verify primary source] (arXiv) — Geometrically and semantically decoupled 4DGS for surgical scene understanding; HexPlane spatiotemporal entanglement module synchronizes semantic changes with scene dynamics - *

- **DeGenseGS** [no arXiv ID; verify primary source] (arXiv) — Geometrically and semantically decoupled 4DGS for surgical scene understanding; HexPlane spatiotemporal entanglement module synchronizes semantic changes with scene dynamics - *

- **DN-4DGS** [arXiv:2410.13607](https://arxiv.org/abs/2410.13607) (NeurIPS 2024) — Denoised deformable network with temporal-spatial aggregation for dynamic scene rendering

- **DN-4DGS** [arXiv:2410.13607](https://arxiv.org/abs/2410.13607) (NeurIPS 2024) — Denoised deformable network with temporal-spatial aggregation for dynamic scene rendering

- **DreamGaussian4D** [arXiv:2312.17142](https://arxiv.org/abs/2312.17142) (arXiv preprint) — Efficient 4D generation with hybrid explicit-implicit Gaussian representations, motion modeling via keyframe interpolation, and DreamFlow for text

- **DreamGaussian4D** [arXiv:2312.17142](https://arxiv.org/abs/2312.17142) (arXiv preprint) — Efficient 4D generation with hybrid explicit-implicit Gaussian representations, motion modeling via keyframe interpolation, and DreamFlow for text

- **DreamMesh4D** [arXiv:2410.06756](https://arxiv.org/abs/2410.06756) (NeurIPS 2024) — Sparse-controlled Gaussian-Mesh hybrid 4D generation

- **DreamMesh4D** [arXiv:2410.06756](https://arxiv.org/abs/2410.06756) (NeurIPS 2024) — Sparse-controlled Gaussian-Mesh hybrid 4D generation

- **Dual-GS** [arXiv:2409.08353](https://arxiv.org/abs/2409.08353) (ACM ToG 2024) — Dual-stream Gaussian Splatting; separates global and local geometry streams for stable reconstruction of long-duration dynamic scenes; dual-stream for long-

- **Dual-GS** [arXiv:2409.08353](https://arxiv.org/abs/2409.08353) (ACM ToG 2024) — Dual-stream Gaussian Splatting; separates global and local geometry streams for stable reconstruction of long-duration dynamic scenes; dual-stream for long-

- **DUET4D** [arXiv:2610.35002](https://arxiv.org/abs/2610.35002) (arXiv 2026) — Self-corrective coupled guidance for sparse-view 4D Gaussian splatting; the 4D representation guides the generative prior toward spatiotemporally consistent ps

- **DUET4D** [arXiv:2610.35002](https://arxiv.org/abs/2610.35002) (arXiv 2026) — Self-corrective coupled guidance for sparse-view 4D Gaussian splatting; the 4D representation guides the generative prior toward spatiotemporally consistent ps

- **DynActiveGS** [arXiv:2608.01178](https://arxiv.org/abs/2608.01178) (arXiv 2026) — Dynamic-aware active reconstruction framework based on 3D Gaussian Splatting (3DGS) for autonomous exploration in dynamic environments

- **DynActiveGS** [arXiv:2608.01178](https://arxiv.org/abs/2608.01178) (arXiv 2026) — Dynamic-aware active reconstruction framework based on 3D Gaussian Splatting (3DGS) for autonomous exploration in dynamic environments

- **DynMF** [arXiv:2312.00112](https://arxiv.org/abs/2312.00112) (CVPR 2024) — Dynamic neural motion fields decomposing scene motion into compact basis functions for 4D GS

- **DynMF** [arXiv:2312.00112](https://arxiv.org/abs/2312.00112) (CVPR 2024) — Dynamic neural motion fields decomposing scene motion into compact basis functions for 4D GS

- **DynOMo** [arXiv:2409.02104](https://arxiv.org/abs/2409.02104) (3DV 2025) — Online motion adaptation for dynamic Gaussian Splatting; adaptive per-frame Gaussian motion estimation handles changing scene dynamics; online motion adaptation 

- **DynOMo** [arXiv:2409.02104](https://arxiv.org/abs/2409.02104) (3DV 2025) — Online motion adaptation for dynamic Gaussian Splatting; adaptive per-frame Gaussian motion estimation handles changing scene dynamics; online motion adaptation 

- **EdMCGS** [arXiv:2609.08332](https://arxiv.org/abs/2609.08332) (arXiv 2026) — Event-driven Markov chain 3DGS for extreme-low-frame-rate dynamic scene reconstruction from RGB + event stream

- **EdMCGS** [arXiv:2609.08332](https://arxiv.org/abs/2609.08332) (arXiv 2026) — Event-driven Markov chain 3DGS for extreme-low-frame-rate dynamic scene reconstruction from RGB + event stream

- **ERF-GS** [arXiv:2608.08531](https://arxiv.org/abs/2608.08531) (arXiv 2026) — Event-RGB fusion Gaussian splatting (ERF-GS) framework that integrates event information into both optimization and densification stages of the Gaussian splatt

- **ERF-GS** [arXiv:2608.08531](https://arxiv.org/abs/2608.08531) (arXiv 2026) — Event-RGB fusion Gaussian splatting (ERF-GS) framework that integrates event information into both optimization and densification stages of the Gaussian splatt

- **FastFlowGS** [arXiv:2609.16310](https://arxiv.org/abs/2609.16310) (arXiv 2026) — Streaming 4DGS for fast-moving subjects from sparse external cameras; fuses sparse matches + optical flow via Kalman temporal update; +12.6% VMAF on Panopt

- **FastFlowGS** [arXiv:2609.16310](https://arxiv.org/abs/2609.16310) (arXiv 2026) — Streaming 4DGS for fast-moving subjects from sparse external cameras; fuses sparse matches + optical flow via Kalman temporal update; +12.6% VMAF on Panopt

- **FreeForm** [arXiv:2605.29318](https://arxiv.org/abs/2605.29318) (CVPR 2026) — Reduced-order deformable simulation from particle-based skinning eigenmodes; 40x faster

- **FreeForm** [arXiv:2605.29318](https://arxiv.org/abs/2605.29318) (CVPR 2026) — Reduced-order deformable simulation from particle-based skinning eigenmodes; 40x faster

- **FreeTimeGS++** [arXiv:2605.03337](https://arxiv.org/abs/2605.03337) (arXiv) — Principled analysis + gated marginalization for 4DGS stability

- **FreeTimeGS++** [arXiv:2605.03337](https://arxiv.org/abs/2605.03337) (arXiv) — Principled analysis + gated marginalization for 4DGS stability

- **FRUC** [no arXiv ID; verify primary source] (arXiv) — Feed-forward 3DGS for dynamic scene reconstruction from uncalibrated collaborative driving views; ego-centric causal occlusion field + zero-initialized injection - **GauFRE** [arXiv:

- **FRUC** [no arXiv ID; verify primary source] (arXiv) — Feed-forward 3DGS for dynamic scene reconstruction from uncalibrated collaborative driving views; ego-centric causal occlusion field + zero-initialized injection - **GauFRE** [arXiv:

- **GauFRE** [arXiv:2312.11458](https://arxiv.org/abs/2312.11458) (WACV 2025) — Gaussian Frequency-domain Radial basis Expansion; radial basis function networks for dynamic Gaussian deformation in frequency domain; evaluated on D-NeRF, Hype

- **GauFRE** [arXiv:2312.11458](https://arxiv.org/abs/2312.11458) (WACV 2025) — Gaussian Frequency-domain Radial basis Expansion; radial basis function networks for dynamic Gaussian deformation in frequency domain; evaluated on D-NeRF, Hype

- **GaussianFlow** [arXiv:2403.12365](https://arxiv.org/abs/2403.12365) (CVPR 2025) — Optical flow-guided 4DGS for temporally consistent dynamic scene reconstruction

- **GaussianFlow** [arXiv:2403.12365](https://arxiv.org/abs/2403.12365) (CVPR 2025) — Optical flow-guided 4DGS for temporally consistent dynamic scene reconstruction

- **GaussianSpline** [arXiv:2610.35224](https://arxiv.org/abs/2610.35224) (arXiv 2026) — Dynamic 4D Gaussian Splatting method that models per-Gaussian motion using adaptive, non-uniform piecewise cubic Bézier splines

- **GaussianSpline** [arXiv:2610.35224](https://arxiv.org/abs/2610.35224) (arXiv 2026) — Dynamic 4D Gaussian Splatting method that models per-Gaussian motion using adaptive, non-uniform piecewise cubic Bézier splines

- **GaussianWorld** [arXiv:2412.10373](https://arxiv.org/abs/2412.10373) (CVPR 2025) — World-model GS: dynamic scene forecasting with Gaussian-based future prediction

- **GaussianWorld** [arXiv:2412.10373](https://arxiv.org/abs/2412.10373) (CVPR 2025) — World-model GS: dynamic scene forecasting with Gaussian-based future prediction

- **GeoRect4D** [arXiv:2604.20784](https://arxiv.org/abs/2604.20784) (arXiv) — Reconstructing dynamic 3D scenes from sparse multi-view videos is highly ill-posed, often leading to geometric collapse, trajectory drift, and floating artifacts

- **GeoRect4D** [arXiv:2604.20784](https://arxiv.org/abs/2604.20784) (arXiv) — Reconstructing dynamic 3D scenes from sparse multi-view videos is highly ill-posed, often leading to geometric collapse, trajectory drift, and floating artifacts

- **GFlow** [arXiv:2405.18426](https://arxiv.org/abs/2405.18426) (AAAI 2025) — Gaussian Flow for dynamic scene reconstruction; flow-based temporal motion modeling with warp refinement for continuous-time rendering; evaluated on D-NeRF (PSNR

- **GFlow** [arXiv:2405.18426](https://arxiv.org/abs/2405.18426) (AAAI 2025) — Gaussian Flow for dynamic scene reconstruction; flow-based temporal motion modeling with warp refinement for continuous-time rendering; evaluated on D-NeRF (PSNR

- **GPS-Gaussian** [arXiv:2312.02155](https://arxiv.org/abs/2312.02155) (CVPR 2024) — Generalizable Pixel-aligned Gaussian Splatting for novel view synthesis of dynamic scenes; feed-forward pipeline without per-scene optimization; generaliz

- **GPS-Gaussian** [arXiv:2312.02155](https://arxiv.org/abs/2312.02155) (CVPR 2024) — Generalizable Pixel-aligned Gaussian Splatting for novel view synthesis of dynamic scenes; feed-forward pipeline without per-scene optimization; generaliz

- **GrainGS** [arXiv:2607.21448](https://arxiv.org/abs/2607.21448) (arXiv 2026) — Hierarchical anchor skeleton + per-Gaussian deformation with stop-gradient; 36.98 dB PSNR, 435.6 FPS, 4.67 MB

- **GrainGS** [arXiv:2607.21448](https://arxiv.org/abs/2607.21448) (arXiv 2026) — Hierarchical anchor skeleton + per-Gaussian deformation with stop-gradient; 36.98 dB PSNR, 435.6 FPS, 4.67 MB

- **Grassmannian Splatting** [arXiv:2607.10489](https://arxiv.org/abs/2607.10489) (arXiv) — Rank-2 spacetime surfels via Grassmannian parameterization; closed-form motion model without deformation field; fastest among compared methods (4.9-

- **Grassmannian Splatting** [arXiv:2607.10489](https://arxiv.org/abs/2607.10489) (arXiv) — Rank-2 spacetime surfels via Grassmannian parameterization; closed-form motion model without deformation field; fastest among compared methods (4.9-

- **Grid4D** [arXiv:2410.20815](https://arxiv.org/abs/2410.20815) (NeurIPS 2024) — 4D decomposed hash encoding for efficient spatiotemporal Gaussian queries in dynamic GS

- **Grid4D** [arXiv:2410.20815](https://arxiv.org/abs/2410.20815) (NeurIPS 2024) — 4D decomposed hash encoding for efficient spatiotemporal Gaussian queries in dynamic GS

- **H-2DGS** [arXiv:2610.30910](https://arxiv.org/abs/2610.30910) (arXiv 2026) — Height-field 2D Gaussian Splatting concentrating a robust 3DGS reconstruction, fusing its depth into a shared TSDF surface, and representing it with curved sur

- **H-2DGS** [arXiv:2610.30910](https://arxiv.org/abs/2610.30910) (arXiv 2026) — Height-field 2D Gaussian Splatting concentrating a robust 3DGS reconstruction, fusing its depth into a shared TSDF surface, and representing it with curved sur

- **HDR-GS** [arXiv:2405.15125](https://arxiv.org/abs/2405.15125) (NeurIPS 2024) — HDR-specific GS luminance encoding + fast tonemapping for 1000x HDR view synthesis

- **HDR-GS** [arXiv:2405.15125](https://arxiv.org/abs/2405.15125) (NeurIPS 2024) — HDR-specific GS luminance encoding + fast tonemapping for 1000x HDR view synthesis

- **HDR-NSFF** [arXiv:2603.08313](https://arxiv.org/abs/2603.08313) (ICLR 2026) — Dynamic HDR radiance fields from alternating-exposure video

- **HDR-NSFF** [arXiv:2603.08313](https://arxiv.org/abs/2603.08313) (ICLR 2026) — Dynamic HDR radiance fields from alternating-exposure video

- **HiCoM** [arXiv:2411.07541](https://arxiv.org/abs/2411.07541) (NeurIPS 2024) — Hierarchical coherent motion for streamable dynamic scene with 3DGS

- **HiCoM** [arXiv:2411.07541](https://arxiv.org/abs/2411.07541) (NeurIPS 2024) — Hierarchical coherent motion for streamable dynamic scene with 3DGS

- **IGS** [arXiv:2503.16979](https://arxiv.org/abs/2503.16979) (CVPR 2025) — Incremental Gaussian Splatting; incremental reconstruction pipeline with per-frame Gaussian management for streaming dynamic scenes; incremental Gaussian managemen

- **IGS** [arXiv:2503.16979](https://arxiv.org/abs/2503.16979) (CVPR 2025) — Incremental Gaussian Splatting; incremental reconstruction pipeline with per-frame Gaussian management for streaming dynamic scenes; incremental Gaussian managemen

- **L2D2-GS** [arXiv:2606.29374](https://arxiv.org/abs/2606.29374) (arXiv 2026) — This paper presents L2D2-GS, a unified framework that reformulates generalizable reconstruction not as a one-shot regression, but as a robust iterative proces

- **L2D2-GS** [arXiv:2606.29374](https://arxiv.org/abs/2606.29374) (arXiv 2026) — This paper presents L2D2-GS, a unified framework that reformulates generalizable reconstruction not as a one-shot regression, but as a robust iterative proces

- **L4GM** [arXiv:2406.10324](https://arxiv.org/abs/2406.10324) (NeurIPS 2024) — Large-scale feed-forward 4D Gaussian reconstruction from video

- **L4GM** [arXiv:2406.10324](https://arxiv.org/abs/2406.10324) (NeurIPS 2024) — Large-scale feed-forward 4D Gaussian reconstruction from video

- **LagrangeGS** [arXiv:2608.22773](https://arxiv.org/abs/2608.22773) (arXiv 2026) — Formulates dynamic 3DGS as a non-conservative Lagrangian system

- **LagrangeGS** [arXiv:2608.22773](https://arxiv.org/abs/2608.22773) (arXiv 2026) — Formulates dynamic 3DGS as a non-conservative Lagrangian system

- **LetCamsGo** [arXiv:2606.04593](https://arxiv.org/abs/2606.04593) (CVPR) — 4D reconstruction from sparse dynamic cameras; inter-camera matching + intra-camera tracking

- **LetCamsGo** [arXiv:2606.04593](https://arxiv.org/abs/2606.04593) (CVPR) — 4D reconstruction from sparse dynamic cameras; inter-camera matching + intra-camera tracking

- **LoopGaussian** [arXiv:2404.08966](https://arxiv.org/abs/2404.08966) (ACM MM 2024) — Loop-based motion representation for dynamic 3DGS; decomposes SE(3) transformations into meaningful sub-operations via loop primitives; loop-based motio

- **LoopGaussian** [arXiv:2404.08966](https://arxiv.org/abs/2404.08966) (ACM MM 2024) — Loop-based motion representation for dynamic 3DGS; decomposes SE(3) transformations into meaningful sub-operations via loop primitives; loop-based motio

- **MoDec-GS** [arXiv:2501.03714](https://arxiv.org/abs/2501.03714) (CVPR 2025) — Motion Decomposition Gaussian Splatting; decomposes scene motion into global and local components with separate deformation networks; evaluated on HyperNeRF (

- **MoDec-GS** [arXiv:2501.03714](https://arxiv.org/abs/2501.03714) (CVPR 2025) — Motion Decomposition Gaussian Splatting; decomposes scene motion into global and local components with separate deformation networks; evaluated on HyperNeRF (

- **MoDGS** [arXiv:2406.00434](https://arxiv.org/abs/2406.00434) (ICLR 2025) — Motion-aware 3D Gaussian Splatting for dynamic scenes; explicit motion field modeling with smoothness constraints for high-quality temporal coherence; motion-awa

- **MoDGS** [arXiv:2406.00434](https://arxiv.org/abs/2406.00434) (ICLR 2025) — Motion-aware 3D Gaussian Splatting for dynamic scenes; explicit motion field modeling with smoothness constraints for high-quality temporal coherence; motion-awa

- **MORPHOS** [arXiv:2606.02491](https://arxiv.org/abs/2606.02491) (arXiv 2026) — Autoregressive 4D generation with Temporal Structured Latents (T-SLAT) unifying mesh/Gaussian/radiance field; causal attention ensures temporal consistency

- **MORPHOS** [arXiv:2606.02491](https://arxiv.org/abs/2606.02491) (arXiv 2026) — Autoregressive 4D generation with Temporal Structured Latents (T-SLAT) unifying mesh/Gaussian/radiance field; causal attention ensures temporal consistency

- **MotionDreamer** [arXiv:2405.20155](https://arxiv.org/abs/2405.20155) (arXiv 2026) — Category-agnostic skeletal animation from 2D video; structural-semantic injection maps visual dynamics to heterogeneous joint hierarchies; 20;000+ 3D mo

- **MotionDreamer** [arXiv:2405.20155](https://arxiv.org/abs/2405.20155) (arXiv 2026) — Category-agnostic skeletal animation from 2D video; structural-semantic injection maps visual dynamics to heterogeneous joint hierarchies; 20;000+ 3D mo

- **MotionGS** [arXiv:2410.07707](https://arxiv.org/abs/2410.07707) (NeurIPS 2024) — Explicit motion field guiding deformable 3DGS, decoupling motion from appearance

- **MotionGS** [arXiv:2410.07707](https://arxiv.org/abs/2410.07707) (NeurIPS 2024) — Explicit motion field guiding deformable 3DGS, decoupling motion from appearance

- **Multi4D** [arXiv:2606.22197](https://arxiv.org/abs/2606.22197) (arXiv 2026) — To resolve this, we introduce Multi4D, a framework for high-fidelity dynamic Gaussian Splatting based on multi-level competitive allocation

- **Multi4D** [arXiv:2606.22197](https://arxiv.org/abs/2606.22197) (arXiv 2026) — To resolve this, we introduce Multi4D, a framework for high-fidelity dynamic Gaussian Splatting based on multi-level competitive allocation

- **MVFusion-GS** [no arXiv ID; verify primary source] (arXiv) — Motion-variance guided refinement + MotionFormer temporal attention for dynamic 3DGS; explicit motion awareness improves foreground motion + static background - **NeuroGauss4D

- **MVFusion-GS** [no arXiv ID; verify primary source] (arXiv) — Motion-variance guided refinement + MotionFormer temporal attention for dynamic 3DGS; explicit motion awareness improves foreground motion + static background - **NeuroGauss4D

- **NeuroGauss4D-PCI** [arXiv:2405.14241](https://arxiv.org/abs/2405.14241) (NeurIPS 2024) — 4D neural field + Gaussian deformation fields for point cloud interpolation

- **NeuroGauss4D-PCI** [arXiv:2405.14241](https://arxiv.org/abs/2405.14241) (NeurIPS 2024) — 4D neural field + Gaussian deformation fields for point cloud interpolation

- **NeuROK** [arXiv:2605.30347](https://arxiv.org/abs/2605.30347) (CVPR) — Generative 4D Neural Object Kinematics; learns latent kinematic state for simulative dynamics

- **NeuROK** [arXiv:2605.30347](https://arxiv.org/abs/2605.30347) (CVPR) — Generative 4D Neural Object Kinematics; learns latent kinematic state for simulative dynamics

- **OuroWorld** [arXiv:2610.12461](https://arxiv.org/abs/2610.12461) (arXiv 2026) — Mask-free framework that turns any static 3DGS scene into a 3D cinemagraph with diverse, seamlessly looping motion from any viewpoint; Inconsistency-Robust 

- **OuroWorld** [arXiv:2610.12461](https://arxiv.org/abs/2610.12461) (arXiv 2026) — Mask-free framework that turns any static 3DGS scene into a 3D cinemagraph with diverse, seamlessly looping motion from any viewpoint; Inconsistency-Robust 

- **OVOW** [arXiv:2606.31388](https://arxiv.org/abs/2606.31388) (ECCV 2026) — Monocular video to instance-level 4D mesh scene for physics simulation; unifies static/rigid/non-rigid; gravity/contact/support correction; 120-scene benchmark

- **OVOW** [arXiv:2606.31388](https://arxiv.org/abs/2606.31388) (ECCV 2026) — Monocular video to instance-level 4D mesh scene for physics simulation; unifies static/rigid/non-rigid; gravity/contact/support correction; 120-scene benchmark

- **PaMoSplat** [arXiv:2605.10307](https://arxiv.org/abs/2605.10307) (TCSVT) — Dynamic scene reconstruction represents a fundamental yet demanding challenge in computer vision and robotics. While recent progress in 3DGS-based methods has ad

- **PaMoSplat** [arXiv:2605.10307](https://arxiv.org/abs/2605.10307) (TCSVT) — Dynamic scene reconstruction represents a fundamental yet demanding challenge in computer vision and robotics. While recent progress in 3DGS-based methods has ad

- **ParticleGS** [arXiv:2505.20270](https://arxiv.org/abs/2505.20270) (CVPR 2026 Highlight) — The ability to extrapolate dynamic 3D scenes beyond the observed timeframe is fundamental to advancing physical world understanding and predictive

- **ParticleGS** [arXiv:2505.20270](https://arxiv.org/abs/2505.20270) (CVPR 2026 Highlight) — The ability to extrapolate dynamic 3D scenes beyond the observed timeframe is fundamental to advancing physical world understanding and predictive

- **PD-4DGS** [arXiv:2605.11427](https://arxiv.org/abs/2605.11427) (arXiv) — 4D Gaussian Splatting (4DGS) enables high-quality dynamic novel view synthesis, yet current models remain monolithic bitstreams that clients must download in full 

- **PD-4DGS** [arXiv:2605.11427](https://arxiv.org/abs/2605.11427) (arXiv) — 4D Gaussian Splatting (4DGS) enables high-quality dynamic novel view synthesis, yet current models remain monolithic bitstreams that clients must download in full 

- **PersistGS** [arXiv:2606.03479](https://arxiv.org/abs/2606.03479) (arXiv 2026) — $textbfPersistGS$, a method that restores object permanence during occlusion by coupling differentiable rigid body simulation with 3D Gaussian Splatting

- **PersistGS** [arXiv:2606.03479](https://arxiv.org/abs/2606.03479) (arXiv 2026) — $textbfPersistGS$, a method that restores object permanence during occlusion by coupling differentiable rigid body simulation with 3D Gaussian Splatting

- **PhysGaussian** [arXiv:2311.12198](https://arxiv.org/abs/2311.12198) (CVPR 2024) — Seamless integration of Newtonian dynamics (custom MPM withIncremental potential contact) into 3D Gaussians for high-quality novel motion synthesis, netwo

- **PhysGaussian** [arXiv:2311.12198](https://arxiv.org/abs/2311.12198) (CVPR 2024) — Seamless integration of Newtonian dynamics (custom MPM withIncremental potential contact) into 3D Gaussians for high-quality novel motion synthesis, netwo

- **PVG** [arXiv:2311.18561](https://arxiv.org/abs/2311.18561) (arXiv 2023) — Periodic Vibration Gaussian; models temporal dynamics via periodic vibration functions enabling 4D representation without explicit deformation networks; periodic 

- **PVG** [arXiv:2311.18561](https://arxiv.org/abs/2311.18561) (arXiv 2023) — Periodic Vibration Gaussian; models temporal dynamics via periodic vibration functions enabling 4D representation without explicit deformation networks; periodic 

- **R5DGS** [arXiv:2605.25909](https://arxiv.org/abs/2605.25909) (arXiv) — Identity Encoding + rigid-body centroid constraint for semantic-aware 4DGS; 11 FPS speedup

- **R5DGS** [arXiv:2605.25909](https://arxiv.org/abs/2605.25909) (arXiv) — Identity Encoding + rigid-body centroid constraint for semantic-aware 4DGS; 11 FPS speedup

- **Real-time 4DGS** [arXiv:2310.10642](https://arxiv.org/abs/2310.10642) (ICLR 2024) — Real-time 4D Gaussian Splatting; 4DGS with temporal slicing for real-time rendering of dynamic scenes at interactive frame rates; evaluated on D-NeRF (P

- **Real-time 4DGS** [arXiv:2310.10642](https://arxiv.org/abs/2310.10642) (ICLR 2024) — Real-time 4D Gaussian Splatting; 4DGS with temporal slicing for real-time rendering of dynamic scenes at interactive frame rates; evaluated on D-NeRF (P

- **ReconDreamer++** [arXiv:2503.18438](https://arxiv.org/abs/2503.18438) (arXiv 2025) — Incremental reconstruction and rendering for dynamic driving scenarios; integrates temporal Gaussian management with driving scene expansion; dynamic d

- **ReconDreamer++** [arXiv:2503.18438](https://arxiv.org/abs/2503.18438) (arXiv 2025) — Incremental reconstruction and rendering for dynamic driving scenarios; integrates temporal Gaussian management with driving scene expansion; dynamic d

- **RiGS** [arXiv:2605.23672](https://arxiv.org/abs/2605.23672) (arXiv) — Static/rigid/transient Gaussian decomposition + scene flow guidance for monocular 4DGS

- **RiGS** [arXiv:2605.23672](https://arxiv.org/abs/2605.23672) (arXiv) — Static/rigid/transient Gaussian decomposition + scene flow guidance for monocular 4DGS

- **RobustSplat** [no arXiv ID; verify primary source] (ICCV 2025) — Decouples densification from dynamics for transient-free 3DGS - **SAGD** [arXiv:2401.17857](https://arxiv.org/abs/2401.17857) (ECCV 2024) — Self-supervised articulated Gau

- **RobustSplat** [no arXiv ID; verify primary source] (ICCV 2025) — Decouples densification from dynamics for transient-free 3DGS - **SAGD** [arXiv:2401.17857](https://arxiv.org/abs/2401.17857) (ECCV 2024) — Self-supervised articulated Gau

- **SAGD** [arXiv:2401.17857](https://arxiv.org/abs/2401.17857) (ECCV 2024) — Self-supervised articulated Gaussian discovery: automatic part segmentation + articulation estimation without supervision

- **SAGD** [arXiv:2401.17857](https://arxiv.org/abs/2401.17857) (ECCV 2024) — Self-supervised articulated Gaussian discovery: automatic part segmentation + articulation estimation without supervision

- **SC-GS** [no arXiv ID; verify primary source] (arXiv) — Spatial-temporal compression for dynamic Gaussians - **SemDynReg** [arXiv:2606.28656](https://arxiv.org/abs/2606.28656) (arXiv 2026) — Deformable 3D Gaussian Splatting (3DGS) has em

- **SC-GS** [no arXiv ID; verify primary source] (arXiv) — Spatial-temporal compression for dynamic Gaussians - **SemDynReg** [arXiv:2606.28656](https://arxiv.org/abs/2606.28656) (arXiv 2026) — Deformable 3D Gaussian Splatting (3DGS) has em

- **SemDynReg** [arXiv:2606.28656](https://arxiv.org/abs/2606.28656) (arXiv 2026) — Deformable 3D Gaussian Splatting (3DGS) has emerged as an efficient approach for rendering dynamic scenes in a wide range of 3D applications

- **SemDynReg** [arXiv:2606.28656](https://arxiv.org/abs/2606.28656) (arXiv 2026) — Deformable 3D Gaussian Splatting (3DGS) has emerged as an efficient approach for rendering dynamic scenes in a wide range of 3D applications

- **Shape of Motion** [arXiv:2407.13764](https://arxiv.org/abs/2407.13764) (ICCV 2025) — Monocular dynamic reconstruction via 2D motion + depth priors into a compact SE(3) motion basis deforming a scene of 3D Gaussians; world-frame consiste

- **Shape of Motion** [arXiv:2407.13764](https://arxiv.org/abs/2407.13764) (ICCV 2025) — Monocular dynamic reconstruction via 2D motion + depth priors into a compact SE(3) motion basis deforming a scene of 3D Gaussians; world-frame consiste

- **SMG** [arXiv:2608.31023](https://arxiv.org/abs/2608.31023) (ECCV 2026) — Semantic Motion Graph for monocular dynamic Gaussian splatting; semantic-motion coupling enables coherent decomposition of complex non-rigid scenes into motion gro

- **SMG** [arXiv:2608.31023](https://arxiv.org/abs/2608.31023) (ECCV 2026) — Semantic Motion Graph for monocular dynamic Gaussian splatting; semantic-motion coupling enables coherent decomposition of complex non-rigid scenes into motion gro

- **SpectroMotion** [arXiv:2410.17249](https://arxiv.org/abs/2410.17249) (CVPR 2025) — Spectral temporal analysis for motion-aware 3DGS; frequency-domain decomposition of Gaussian motions handles both slow and fast dynamics; evaluated on Hy

- **SpectroMotion** [arXiv:2410.17249](https://arxiv.org/abs/2410.17249) (CVPR 2025) — Spectral temporal analysis for motion-aware 3DGS; frequency-domain decomposition of Gaussian motions handles both slow and fast dynamics; evaluated on Hy

- **SP-GS** [arXiv:2406.03697](https://arxiv.org/abs/2406.03697) (ICML 2024) — Gaussian Splatting with Sparse Pixels; sparse-pixel control points drive deformation field for efficient dynamic scene reconstruction; evaluated on D-NeRF (PSNR 

- **SP-GS** [arXiv:2406.03697](https://arxiv.org/abs/2406.03697) (ICML 2024) — Gaussian Splatting with Sparse Pixels; sparse-pixel control points drive deformation field for efficient dynamic scene reconstruction; evaluated on D-NeRF (PSNR 

- **SplineGS** [arXiv:2412.09982](https://arxiv.org/abs/2412.09982) (CVPR 2025) — Spline-based sparse control for dynamic Gaussian Splatting; B-spline trajectories of sparse control points model continuous scene deformation; B-spline sparse

- **SplineGS** [arXiv:2412.09982](https://arxiv.org/abs/2412.09982) (CVPR 2025) — Spline-based sparse control for dynamic Gaussian Splatting; B-spline trajectories of sparse control points model continuous scene deformation; B-spline sparse

- **STC-GS** [arXiv:2502.14895](https://arxiv.org/abs/2502.14895) (ICLR 2025) — SpatioTemporal Coherent Gaussian Splatting for 3D radar sequence representation in weather nowcasting: per-frame 3D Gaussian groups with cross-frame consistent 

- **STC-GS** [arXiv:2502.14895](https://arxiv.org/abs/2502.14895) (ICLR 2025) — SpatioTemporal Coherent Gaussian Splatting for 3D radar sequence representation in weather nowcasting: per-frame 3D Gaussian groups with cross-frame consistent 

- **STG** [arXiv:2510.22140](https://arxiv.org/abs/2510.22140) (CVPR 2025) — Spatiotemporal Gaussians with decomposed spatial-temporal attention for 4D rendering

- **STG** [arXiv:2510.22140](https://arxiv.org/abs/2510.22140) (CVPR 2025) — Spatiotemporal Gaussians with decomposed spatial-temporal attention for 4D rendering

- **SwitchSplat** [arXiv:2610.24812](https://arxiv.org/abs/2610.24812) (arXiv preprint) — Abstract context: 3D Gaussian Splatting has made high-fidelity 3D reconstruction fast and accessible, yet existing representations assume a static wor

- **SwitchSplat** [arXiv:2610.24812](https://arxiv.org/abs/2610.24812) (arXiv preprint) — Abstract context: 3D Gaussian Splatting has made high-fidelity 3D reconstruction fast and accessible, yet existing representations assume a static wor

- **T2Mo** [arXiv:2606.05162](https://arxiv.org/abs/2606.05162) (arXiv) — Controllable dynamic 3D shape generation via 3D trajectories + text; shape-grounded trajectory embedding

- **T2Mo** [arXiv:2606.05162](https://arxiv.org/abs/2606.05162) (arXiv) — Controllable dynamic 3D shape generation via 3D trajectories + text; shape-grounded trajectory embedding

- **TC4D** [arXiv:2403.17920](https://arxiv.org/abs/2403.17920) (ECCV 2024) — Trajectory-conditioned text-to-4D generation with motion amplitude scheduling and spatial-temporal consistency from a text-to-video model; hybrid explicit-implici

- **TC4D** [arXiv:2403.17920](https://arxiv.org/abs/2403.17920) (ECCV 2024) — Trajectory-conditioned text-to-4D generation with motion amplitude scheduling and spatial-temporal consistency from a text-to-video model; hybrid explicit-implici

- **TIDES** [arXiv:2606.02058](https://arxiv.org/abs/2606.02058) (arXiv 2026) — Continuous-time event camera simulator from dynamic Gaussian Splatting; derives per-pixel intensity dynamics from scene model (not frame differencing); multi-th

- **TIDES** [arXiv:2606.02058](https://arxiv.org/abs/2606.02058) (arXiv 2026) — Continuous-time event camera simulator from dynamic Gaussian Splatting; derives per-pixel intensity dynamics from scene model (not frame differencing); multi-th

- **UniFusion** [arXiv:2609.05888](https://arxiv.org/abs/2609.05888) (ECCV 2026) — Unified spatio-temporal depth alignment for sparse-view 4D reconstruction; spatio-temporal neural fields align cross-view/cross-time depth without masks

- **UniFusion** [arXiv:2609.05888](https://arxiv.org/abs/2609.05888) (ECCV 2026) — Unified spatio-temporal depth alignment for sparse-view 4D reconstruction; spatio-temporal neural fields align cross-view/cross-time depth without masks

- **Velox** [arXiv:2605.04527](https://arxiv.org/abs/2605.04527) (arXiv 2026) — We introduce a framework for learning latent representations of 4D objects which are descriptive, faithfully capturing object geometry and appearance; compressi

- **Velox** [arXiv:2605.04527](https://arxiv.org/abs/2605.04527) (arXiv 2026) — We introduce a framework for learning latent representations of 4D objects which are descriptive, faithfully capturing object geometry and appearance; compressi

- **Video-3DGS** [arXiv:2406.02541](https://arxiv.org/abs/2406.02541) (arXiv 2024) — Video-based 3D Gaussian Splatting; per-frame Gaussian initialization with inter-frame motion priors for video-to-3D dynamic reconstruction; video-based per

- **Video-3DGS** [arXiv:2406.02541](https://arxiv.org/abs/2406.02541) (arXiv 2024) — Video-based 3D Gaussian Splatting; per-frame Gaussian initialization with inter-frame motion priors for video-to-3D dynamic reconstruction; video-based per

- **Vidu4D** [arXiv:2405.16822](https://arxiv.org/abs/2405.16822) (NeurIPS 2024) — Dynamic Gaussian Surfels for single-video to 4D reconstruction

- **Vidu4D** [arXiv:2405.16822](https://arxiv.org/abs/2405.16822) (NeurIPS 2024) — Dynamic Gaussian Surfels for single-video to 4D reconstruction

- **WebSpline** [arXiv:2606.02096](https://arxiv.org/abs/2606.02096) (arXiv 2026) — Structure-Informed Spline (learnable Hermite spline) models Gaussian trajectories + Structural Proxy Graph organizes motion; SOTA on iPhone dataset

- **WebSpline** [arXiv:2606.02096](https://arxiv.org/abs/2606.02096) (arXiv 2026) — Structure-Informed Spline (learnable Hermite spline) models Gaussian trajectories + Structural Proxy Graph organizes motion; SOTA on iPhone dataset

- **3DGS-Avatar** [arXiv:2312.09228](https://arxiv.org/abs/2312.09228) (CVPR 2024) — Deformable 3DGS for animatable human avatars with pose-conditioned Gaussian deformation

- **3DGS-Avatar** [arXiv:2312.09228](https://arxiv.org/abs/2312.09228) (CVPR 2024) — Deformable 3DGS for animatable human avatars with pose-conditioned Gaussian deformation

- **ArtMesh** [arXiv:2605.16582](https://arxiv.org/abs/2605.16582) (arXiv 2026) — Part-aware articulated mesh field with restricted Delaunay remeshing + bidirectional vertex-wise motion consistency

- **ArtMesh** [arXiv:2605.16582](https://arxiv.org/abs/2605.16582) (arXiv 2026) — Part-aware articulated mesh field with restricted Delaunay remeshing + bidirectional vertex-wise motion consistency

- **CapTalk** [arXiv:2605.29316](https://arxiv.org/abs/2605.29316) (arXiv) — Text-guided stylization and speech-driven 3D head animation; separate control over style and emotion; dynamic emotion during inference

- **CapTalk** [arXiv:2605.29316](https://arxiv.org/abs/2605.29316) (arXiv) — Text-guided stylization and speech-driven 3D head animation; separate control over style and emotion; dynamic emotion during inference

- **CGR-GS** [arXiv:2610.35152](https://arxiv.org/abs/2610.35152) (arXiv 2026) — Complexity-guided regularization for generalizable human Gaussian splatting; local-entropy complexity priors prevent coarse Gaussian placement in complex regio

- **CGR-GS** [arXiv:2610.35152](https://arxiv.org/abs/2610.35152) (arXiv 2026) — Complexity-guided regularization for generalizable human Gaussian splatting; local-entropy complexity priors prevent coarse Gaussian placement in complex regio

- **COSY** [arXiv:2605.24114](https://arxiv.org/abs/2605.24114) (arXiv) — Compositional GAN generates hair/skin/glasses/torso independently for disentangled head editing

- **COSY** [arXiv:2605.24114](https://arxiv.org/abs/2605.24114) (arXiv) — Compositional GAN generates hair/skin/glasses/torso independently for disentangled head editing

- **D-Rex** [arXiv:2604.27871](https://arxiv.org/abs/2604.27871) (SIGGRAPH 2026) — Diffusion post-process relighting for expressive avatars

- **D-Rex** [arXiv:2604.27871](https://arxiv.org/abs/2604.27871) (SIGGRAPH 2026) — Diffusion post-process relighting for expressive avatars

- **EGGHead** [arXiv:2610.26898](https://arxiv.org/abs/2610.26898) (arXiv preprint) — Abstract context: Creating photorealistic human heads typically involves trade-offs between representation power, efficiency, and generalizability.

- **EGGHead** [arXiv:2610.26898](https://arxiv.org/abs/2610.26898) (arXiv preprint) — Abstract context: Creating photorealistic human heads typically involves trade-offs between representation power, efficiency, and generalizability.

- **EmoZone-Talker** [arXiv:2606.15848](https://arxiv.org/abs/2606.15848) (arXiv 2026) — Novel framework that reformulates audio-driven facial animation as a structured spatial-temporal coordination problem under cross-modal conflicts

- **EmoZone-Talker** [arXiv:2606.15848](https://arxiv.org/abs/2606.15848) (arXiv 2026) — Novel framework that reformulates audio-driven facial animation as a structured spatial-temporal coordination problem under cross-modal conflicts

- **ExpressiveGaussianHuman** [arXiv:2407.03204](https://arxiv.org/abs/2407.03204) (NeurIPS 2024) — Expression-coefficient-driven Gaussian deformation fields for expressive human avatars

- **ExpressiveGaussianHuman** [arXiv:2407.03204](https://arxiv.org/abs/2407.03204) (NeurIPS 2024) — Expression-coefficient-driven Gaussian deformation fields for expressive human avatars

- **FlexiAvatar** [arXiv:2607.19100](https://arxiv.org/abs/2607.19100) (ECCV 2026) — Unified framework optimizing only visible body regions; SMPL-X tracking + diffusion completion

- **FlexiAvatar** [arXiv:2607.19100](https://arxiv.org/abs/2607.19100) (ECCV 2026) — Unified framework optimizing only visible body regions; SMPL-X tracking + diffusion completion

- **GAGAvatar** [arXiv:2410.07971](https://arxiv.org/abs/2410.07971) (NeurIPS 2024) — Generalizable and animatable Gaussian head avatar from monocular video

- **GAGAvatar** [arXiv:2410.07971](https://arxiv.org/abs/2410.07971) (NeurIPS 2024) — Generalizable and animatable Gaussian head avatar from monocular video

- **GAS** [no arXiv ID; verify primary source] (arXiv) — Compression + caching for real-time avatar rendering - **GauHuman** [arXiv:2312.02973](https://arxiv.org/abs/2312.02973) (ECCV 2024) — Human-specific GS with SMPL-constrained Gaussian

- **GAS** [no arXiv ID; verify primary source] (arXiv) — Compression + caching for real-time avatar rendering - **GauHuman** [arXiv:2312.02973](https://arxiv.org/abs/2312.02973) (ECCV 2024) — Human-specific GS with SMPL-constrained Gaussian

- **GauHuman** [arXiv:2312.02973](https://arxiv.org/abs/2312.02973) (ECCV 2024) — Human-specific GS with SMPL-constrained Gaussian initialization and pose-aware densification

- **GauHuman** [arXiv:2312.02973](https://arxiv.org/abs/2312.02973) (ECCV 2024) — Human-specific GS with SMPL-constrained Gaussian initialization and pose-aware densification

- **GaussianAvatar** [no arXiv ID; verify primary source] (arXiv) — Pose-driven human body Gaussian representation - **GaussianBody** [arXiv:2401.09720](https://arxiv.org/abs/2401.09720) (CVPR 2025) — SMPL-X aligned Gaussian body with part-

- **GaussianAvatar** [no arXiv ID; verify primary source] (arXiv) — Pose-driven human body Gaussian representation - **GaussianBody** [arXiv:2401.09720](https://arxiv.org/abs/2401.09720) (CVPR 2025) — SMPL-X aligned Gaussian body with part-

- **GaussianBody** [arXiv:2401.09720](https://arxiv.org/abs/2401.09720) (CVPR 2025) — SMPL-X aligned Gaussian body with part-aware densification

- **GaussianBody** [arXiv:2401.09720](https://arxiv.org/abs/2401.09720) (CVPR 2025) — SMPL-X aligned Gaussian body with part-aware densification

- **GaussianEmoTalker** [arXiv:2607.00959](https://arxiv.org/abs/2607.00959) (arXiv 2026) — Audio-driven framework for real-time emotional talking head synthesis based on 3D Gaussian Splatting

- **GaussianEmoTalker** [arXiv:2607.00959](https://arxiv.org/abs/2607.00959) (arXiv 2026) — Audio-driven framework for real-time emotional talking head synthesis based on 3D Gaussian Splatting

- **GaussianTalker** [arXiv:2404.16012](https://arxiv.org/abs/2404.16012) (CVPR 2025) — Audio-driven Gaussian talking head with facial prior and emotion control

- **GaussianTalker** [arXiv:2404.16012](https://arxiv.org/abs/2404.16012) (CVPR 2025) — Audio-driven Gaussian talking head with facial prior and emotion control

- **GradRig** [arXiv:2609.05127](https://arxiv.org/abs/2609.05127) (arXiv 2026) — Differentiable weights for skinned Gaussian splat deformation; spatial gradient of skinning weights for mesh-free deformation

- **GradRig** [arXiv:2609.05127](https://arxiv.org/abs/2609.05127) (arXiv 2026) — Differentiable weights for skinned Gaussian splat deformation; spatial gradient of skinning weights for mesh-free deformation

- **G-Skin** [arXiv:2608.01726](https://arxiv.org/abs/2608.01726) (arXiv 2026) — To address this challenging problem, we propose G-Skin, a novel generative skinning framework designed for expressive and high-fidelity animation with 3D Gauss

- **G-Skin** [arXiv:2608.01726](https://arxiv.org/abs/2608.01726) (arXiv 2026) — To address this challenging problem, we propose G-Skin, a novel generative skinning framework designed for expressive and high-fidelity animation with 3D Gauss

- **HairGPT** [arXiv:2605.08824](https://arxiv.org/abs/2605.08824) (SIGGRAPH 2026 Journal Track) — Strand-as-Language autoregressive modeling for 3D hairstyle synthesis

- **HairGPT** [arXiv:2605.08824](https://arxiv.org/abs/2605.08824) (SIGGRAPH 2026 Journal Track) — Strand-as-Language autoregressive modeling for 3D hairstyle synthesis

- **Hand-4DGS** [arXiv:2606.19156](https://arxiv.org/abs/2606.19156) (arXiv 2026) — To address these challenges, we introduce Hand-4DGS, the first feed-forward framework for reconstructing dynamic 4D hands directly from egocentric videos, e

- **Hand-4DGS** [arXiv:2606.19156](https://arxiv.org/abs/2606.19156) (arXiv 2026) — To address these challenges, we introduce Hand-4DGS, the first feed-forward framework for reconstructing dynamic 4D hands directly from egocentric videos, e

- **HeadGaS** [arXiv:2312.02902](https://arxiv.org/abs/2312.02902) (ECCV 2024) — Dynamic head GS with blendshape-driven Gaussian deformation for real-time reenactment

- **HeadGaS** [arXiv:2312.02902](https://arxiv.org/abs/2312.02902) (ECCV 2024) — Dynamic head GS with blendshape-driven Gaussian deformation for real-time reenactment

- **HeadsUp** [arXiv:2605.04035](https://arxiv.org/abs/2605.04035) (ECCV 2026) — UV-parameterized 3D Gaussian representation decoupling primitive count from input resolution; trained on 10,000+ subjects for large-scale, feed-forward 3D Gaus

- **HeadsUp** [arXiv:2605.04035](https://arxiv.org/abs/2605.04035) (ECCV 2026) — UV-parameterized 3D Gaussian representation decoupling primitive count from input resolution; trained on 10,000+ subjects for large-scale, feed-forward 3D Gaus

- **High-Fidelity Human GS** [arXiv:2604.21714](https://arxiv.org/abs/2604.21714) (arXiv) — SMPL-X geometric priors + region-aware initialization

- **High-Fidelity Human GS** [arXiv:2604.21714](https://arxiv.org/abs/2604.21714) (arXiv) — SMPL-X geometric priors + region-aware initialization

- **Human3Diffusion** [arXiv:2406.08475](https://arxiv.org/abs/2406.08475) (NeurIPS 2024) — Diffusion + GS dual-driven 3D human avatar reconstruction

- **Human3Diffusion** [arXiv:2406.08475](https://arxiv.org/abs/2406.08475) (NeurIPS 2024) — Diffusion + GS dual-driven 3D human avatar reconstruction

- **HumanSplat** [arXiv:2406.12459](https://arxiv.org/abs/2406.12459) (NeurIPS 2024) — Single-image human Gaussian reconstruction combining multi-view diffusion, a latent reconstruction transformer and human-structure priors.

- **HumanSplat** [arXiv:2406.12459](https://arxiv.org/abs/2406.12459) (NeurIPS 2024) — Single-image human Gaussian reconstruction combining multi-view diffusion, a latent reconstruction transformer and human-structure priors.

- **HumanSplatHMR** [arXiv:2605.02784](https://arxiv.org/abs/2605.02784) (arXiv) — Joint pose refinement + Gaussian avatar optimization

- **HumanSplatHMR** [arXiv:2605.02784](https://arxiv.org/abs/2605.02784) (arXiv) — Joint pose refinement + Gaussian avatar optimization

- **Latent Dynamics** [arXiv:2605.21478](https://arxiv.org/abs/2605.21478) (arXiv 2026) — Pose-driven 3DGS avatar with transformer decoder + dynamics residual latent; learned force decomposition (driving/restoring/dissipative) for temporall

- **Latent Dynamics** [arXiv:2605.21478](https://arxiv.org/abs/2605.21478) (arXiv 2026) — Pose-driven 3DGS avatar with transformer decoder + dynamics residual latent; learned force decomposition (driving/restoring/dissipative) for temporall

- **Multi-view Consistent 3D Gaussian Head Avatars** [arXiv:2605.25220](https://arxiv.org/abs/2605.25220) (CVPR) — Multi-view consistent 3D Gaussian head avatars from single-view input without multi-view generation

- **Multi-view Consistent 3D Gaussian Head Avatars** [arXiv:2605.25220](https://arxiv.org/abs/2605.25220) (CVPR) — Multi-view consistent 3D Gaussian head avatars from single-view input without multi-view generation

- **OASIS** [arXiv:2607.29633](https://arxiv.org/abs/2607.29633) (arXiv 2026) — Tailored 3D Gaussian Splatting framework for single-image hand avatar reconstruction

- **OASIS** [arXiv:2607.29633](https://arxiv.org/abs/2607.29633) (arXiv 2026) — Tailored 3D Gaussian Splatting framework for single-image hand avatar reconstruction

- **PD-GS** [arXiv:2608.05218](https://arxiv.org/abs/2608.05218) (arXiv 2026) — TextbfPhoneme-Driven Gaussian Splatting (PD-GS), which augments a 3DGS talker with time-aligned phoneme tokens obtained from an automatic ASR and forced-alignme

- **PD-GS** [arXiv:2608.05218](https://arxiv.org/abs/2608.05218) (arXiv 2026) — TextbfPhoneme-Driven Gaussian Splatting (PD-GS), which augments a 3DGS talker with time-aligned phoneme tokens obtained from an automatic ASR and forced-alignme

- **PEAR** [no arXiv ID; verify primary source] (SIGGRAPH 2026) — Single-image animatable 3D human avatar at 100 FPS; full-body + hand + face reconstruction from single natural image - **PiG-Avatar** [arXiv:2605.20185](https://arxiv.org/abs

- **PEAR** [no arXiv ID; verify primary source] (SIGGRAPH 2026) — Single-image animatable 3D human avatar at 100 FPS; full-body + hand + face reconstruction from single natural image - **PiG-Avatar** [arXiv:2605.20185](https://arxiv.org/abs

- **PiG-Avatar** [arXiv:2605.20185](https://arxiv.org/abs/2605.20185) (arXiv 2026) — Hierarchical neural-field-guided Gaussian avatars in volumetric canonical space decoupled from template topology; 3D barycentric anchor transport for kinem

- **PiG-Avatar** [arXiv:2605.20185](https://arxiv.org/abs/2605.20185) (arXiv 2026) — Hierarchical neural-field-guided Gaussian avatars in volumetric canonical space decoupled from template topology; 3D barycentric anchor transport for kinem

- **ProgressiveAvatars** [no arXiv ID; verify primary source] (SIGGRAPH 2026 TOG) — Progressive animatable 3D Gaussian avatar generation - **RAGA** [arXiv:2606.29329](https://arxiv.org/abs/2606.29329) (arXiv 2026) — RAGA, a Ray-Traced Gauss

- **ProgressiveAvatars** [no arXiv ID; verify primary source] (SIGGRAPH 2026 TOG) — Progressive animatable 3D Gaussian avatar generation - **RAGA** [arXiv:2606.29329](https://arxiv.org/abs/2606.29329) (arXiv 2026) — RAGA, a Ray-Traced Gauss

- **RAGA** [arXiv:2606.29329](https://arxiv.org/abs/2606.29329) (arXiv 2026) — RAGA, a Ray-Traced Gaussian Shadow Casting formulation based on exact ray-Gaussian line integrals

- **RAGA** [arXiv:2606.29329](https://arxiv.org/abs/2606.29329) (arXiv 2026) — RAGA, a Ray-Traced Gaussian Shadow Casting formulation based on exact ray-Gaussian line integrals

- **S-Avatar** [arXiv:2607.28164](https://arxiv.org/abs/2607.28164) (arXiv 2026) — Novel method for generating photorealistic 3D head avatars from a single image using a diffusion-guided 3D model generation module and strategies for animati

- **S-Avatar** [arXiv:2607.28164](https://arxiv.org/abs/2607.28164) (arXiv 2026) — Novel method for generating photorealistic 3D head avatars from a single image using a diffusion-guided 3D model generation module and strategies for animati

- **SDTalk** [arXiv:2605.09956](https://arxiv.org/abs/2605.09956) (arXiv) — High-quality, real-time talking head synthesis remains a fundamental challenge in computer vision. Existing reconstruction- and rendering-based methods typically re

- **SDTalk** [arXiv:2605.09956](https://arxiv.org/abs/2605.09956) (arXiv) — High-quality, real-time talking head synthesis remains a fundamental challenge in computer vision. Existing reconstruction- and rendering-based methods typically re

- **SpatialAvatar-0** [arXiv:2606.15659](https://arxiv.org/abs/2606.15659) (arXiv 2026) — 10K-iter layout-preserving per-subject refinement loop that freezes the FLAME-binding and Gaussian count and replaces densification with a three-compo

- **SpatialAvatar-0** [arXiv:2606.15659](https://arxiv.org/abs/2606.15659) (arXiv 2026) — 10K-iter layout-preserving per-subject refinement loop that freezes the FLAME-binding and Gaussian count and replaces densification with a three-compo

- **SplatArmor** [arXiv:2311.10812](https://arxiv.org/abs/2311.10812) (CVPR 2024) — LBS-based articulated Gaussian Splatting for human body with twist-aware deformation

- **SplatArmor** [arXiv:2311.10812](https://arxiv.org/abs/2311.10812) (CVPR 2024) — LBS-based articulated Gaussian Splatting for human body with twist-aware deformation

- **SplatFace** [arXiv:2403.18784](https://arxiv.org/abs/2403.18784) (CVPR 2025) — Face-specific GS with identity-preserving Gaussian anchoring from single image

- **SplatFace** [arXiv:2403.18784](https://arxiv.org/abs/2403.18784) (CVPR 2025) — Face-specific GS with identity-preserving Gaussian anchoring from single image

- **SplatPose** [arXiv:2503.05174](https://arxiv.org/abs/2503.05174) (CVPR 2025) — Pose-conditioned Gaussian Splatting for monocular human reconstruction

- **SplatPose** [arXiv:2503.05174](https://arxiv.org/abs/2503.05174) (CVPR 2025) — Pose-conditioned Gaussian Splatting for monocular human reconstruction

- **SplatTalk** [arXiv:2503.06271](https://arxiv.org/abs/2503.06271) (CVPR 2025) — Audio-driven 3D talking face GS with emotion and style control

- **SplatTalk** [arXiv:2503.06271](https://arxiv.org/abs/2503.06271) (CVPR 2025) — Audio-driven 3D talking face GS with emotion and style control

- **SplattingAvatar** [no arXiv ID; verify primary source] (arXiv) — Expression-conditioned Gaussian deformation - **SplitAvatar** [arXiv:2605.25751](https://arxiv.org/abs/2605.25751) (arXiv) — Autoregressive Gaussian splitting via GNN for 

- **SplattingAvatar** [no arXiv ID; verify primary source] (arXiv) — Expression-conditioned Gaussian deformation - **SplitAvatar** [arXiv:2605.25751](https://arxiv.org/abs/2605.25751) (arXiv) — Autoregressive Gaussian splitting via GNN for 

- **SplitAvatar** [arXiv:2605.25751](https://arxiv.org/abs/2605.25751) (arXiv) — Autoregressive Gaussian splitting via GNN for one-shot head avatar with progressive detail

- **SplitAvatar** [arXiv:2605.25751](https://arxiv.org/abs/2605.25751) (arXiv) — Autoregressive Gaussian splitting via GNN for one-shot head avatar with progressive detail

- **Video Diffusion Avatars** [arXiv:2610.23772](https://arxiv.org/abs/2610.23772) (arXiv preprint) — Abstract context: Video diffusion models (VDMs) trained on internet-scale data encode a rich prior on appearance, motion, and physics that

- **Video Diffusion Avatars** [arXiv:2610.23772](https://arxiv.org/abs/2610.23772) (arXiv preprint) — Abstract context: Video diffusion models (VDMs) trained on internet-scale data encode a rich prior on appearance, motion, and physics that

- **World2HOSI** [arXiv:2610.27725](https://arxiv.org/abs/2610.27725) (arXiv preprint) — Abstract context: Long-horizon human-object-scene interaction (HOSI) is central to embodied AI, animation, and game development.

- **World2HOSI** [arXiv:2610.27725](https://arxiv.org/abs/2610.27725) (arXiv preprint) — Abstract context: Long-horizon human-object-scene interaction (HOSI) is central to embodied AI, animation, and game development.

- **3DGS-Enhancer** [arXiv:2410.16266](https://arxiv.org/abs/2410.16266) (NeurIPS 2024) — 2D diffusion priors guiding iterative 3DGS refinement for view-consistent enhancement

- **3DGS-Enhancer** [arXiv:2410.16266](https://arxiv.org/abs/2410.16266) (NeurIPS 2024) — 2D diffusion priors guiding iterative 3DGS refinement for view-consistent enhancement

- **ACD-GS** [arXiv:2610.39399](https://arxiv.org/abs/2610.39399) (arXiv 2026) — Asymmetric curvature-aware densification using second-order depth responses as structural prior to regulate cloning and splitting, with multi-view photometric 

- **ACD-GS** [arXiv:2610.39399](https://arxiv.org/abs/2610.39399) (arXiv 2026) — Asymmetric curvature-aware densification using second-order depth responses as structural prior to regulate cloning and splitting, with multi-view photometric 

- **AdaGScale** [arXiv:2604.18980](https://arxiv.org/abs/2604.18980) (DAC 2026) — Reducing the number of Gaussian-tile pairs is one of the most promising approaches to improve 3D Gaussian Splatting (3D-GS) rendering speed on GPUs. However, 

- **AdaGScale** [arXiv:2604.18980](https://arxiv.org/abs/2604.18980) (DAC 2026) — Reducing the number of Gaussian-tile pairs is one of the most promising approaches to improve 3D Gaussian Splatting (3D-GS) rendering speed on GPUs. However, 

- **AdpSplit** [arXiv:2605.06876](https://arxiv.org/abs/2605.06876) (arXiv) — Adaptive density control in 3D Gaussian Splatting (3DGS) repeatedly grows the Gaussian population through fixed-cardinality random splitting to discover useful sc

- **AdpSplit** [arXiv:2605.06876](https://arxiv.org/abs/2605.06876) (arXiv) — Adaptive density control in 3D Gaussian Splatting (3DGS) repeatedly grows the Gaussian population through fixed-cardinality random splitting to discover useful sc

- **ArtifactWorld** [arXiv:2604.12251](https://arxiv.org/abs/2604.12251) (arXiv) — Video generation models for 3DGS artifact restoration at scale

- **ArtifactWorld** [arXiv:2604.12251](https://arxiv.org/abs/2604.12251) (arXiv) — Video generation models for 3DGS artifact restoration at scale

- **AV1-3DGS** [arXiv:2605.14629](https://arxiv.org/abs/2605.14629) (arXiv 2026) — AV1 motion vectors for dense SfM matching with 63% training time reduction

- **AV1-3DGS** [arXiv:2605.14629](https://arxiv.org/abs/2605.14629) (arXiv 2026) — AV1 motion vectors for dense SfM matching with 63% training time reduction

- **CAdam** [arXiv:2605.20872](https://arxiv.org/abs/2605.20872) (SIGGRAPH 2026) — Context-Adaptive Moment Estimation for 3DGS densification in generative distillation; first moment interference principle; 85-97% Gaussian reduction

- **CAdam** [arXiv:2605.20872](https://arxiv.org/abs/2605.20872) (SIGGRAPH 2026) — Context-Adaptive Moment Estimation for 3DGS densification in generative distillation; first moment interference principle; 85-97% Gaussian reduction

- **Compact Neural Appearance** [arXiv:2609.05255](https://arxiv.org/abs/2609.05255) (arXiv 2026) — Compact neural appearance models for efficient 3DGS; shared MLP decodes latent codes, 192->28 bytes per primitive vs SH3

- **Compact Neural Appearance** [arXiv:2609.05255](https://arxiv.org/abs/2609.05255) (arXiv 2026) — Compact neural appearance models for efficient 3DGS; shared MLP decodes latent codes, 192->28 bytes per primitive vs SH3

- **ConFi-GS** [arXiv:2605.24964](https://arxiv.org/abs/2605.24964) (arXiv) — Confidence-guided high-frequency injection + reliability-aware densification for 3DGS super-resolution

- **ConFi-GS** [arXiv:2605.24964](https://arxiv.org/abs/2605.24964) (arXiv) — Confidence-guided high-frequency injection + reliability-aware densification for 3DGS super-resolution

- **Counting Contributions, Not Views** [arXiv:2610.27108](https://arxiv.org/abs/2610.27108) (arXiv preprint) — Abstract context: 3D Gaussian Splatting grows a dense scene from a sparse initialization through adaptive density control (ADC),

- **Counting Contributions, Not Views** [arXiv:2610.27108](https://arxiv.org/abs/2610.27108) (arXiv preprint) — Abstract context: 3D Gaussian Splatting grows a dense scene from a sparse initialization through adaptive density control (ADC),

- **CrossFit-GS** [arXiv:2610.28190](https://arxiv.org/abs/2610.28190) (arXiv preprint) — Abstract context: Pose-free 3D Gaussian Splatting (3DGS) aims to bypass cumbersome Structure-from-Motion (SfM) by directly reconstructing scenes from 

- **CrossFit-GS** [arXiv:2610.28190](https://arxiv.org/abs/2610.28190) (arXiv preprint) — Abstract context: Pose-free 3D Gaussian Splatting (3DGS) aims to bypass cumbersome Structure-from-Motion (SfM) by directly reconstructing scenes from 

- **DC-Gaussian** [arXiv:2405.17705](https://arxiv.org/abs/2405.17705) (NeurIPS 2024) — Reflection separation + degradation-aware training for reflective dashcam 3DGS

- **DC-Gaussian** [arXiv:2405.17705](https://arxiv.org/abs/2405.17705) (NeurIPS 2024) — Reflection separation + degradation-aware training for reflective dashcam 3DGS

- **Denoising-GS** [arXiv:2605.14880](https://arxiv.org/abs/2605.14880) (arXiv) — Recent advances in 3D Gaussian Splatting (3DGS) have achieved remarkable success in high-fidelity Novel View Synthesis (NVS), yet the optimization process ine

- **Denoising-GS** [arXiv:2605.14880](https://arxiv.org/abs/2605.14880) (arXiv) — Recent advances in 3D Gaussian Splatting (3DGS) have achieved remarkable success in high-fidelity Novel View Synthesis (NVS), yet the optimization process ine

- **DP-Splat** [arXiv:2607.10912](https://arxiv.org/abs/2607.10912) (arXiv) — Bayesian nonparametric complexity control via Dirichlet-process prior; 5.9-7.6x fewer components at matched quality; exact monotonicity guarantee + truncation-err

- **DP-Splat** [arXiv:2607.10912](https://arxiv.org/abs/2607.10912) (arXiv) — Bayesian nonparametric complexity control via Dirichlet-process prior; 5.9-7.6x fewer components at matched quality; exact monotonicity guarantee + truncation-err

- **DualSplat** [arXiv:2604.21631](https://arxiv.org/abs/2604.21631) (CVPR 2026) — Failure-to-Prior framework from reconstruction failures

- **DualSplat** [arXiv:2604.21631](https://arxiv.org/abs/2604.21631) (CVPR 2026) — Failure-to-Prior framework from reconstruction failures

- **E2EGS** [arXiv:2603.14684](https://arxiv.org/abs/2603.14684) (CVPR 2026) — Event-to-edge pose-free 3D reconstruction

- **E2EGS** [arXiv:2603.14684](https://arxiv.org/abs/2603.14684) (CVPR 2026) — Event-to-edge pose-free 3D reconstruction

- **ELoG-GS** [arXiv:2604.12592](https://arxiv.org/abs/2604.12592) (arXiv) — Dual-branch luminance-guided extreme low-light GS enhancement

- **ELoG-GS** [arXiv:2604.12592](https://arxiv.org/abs/2604.12592) (arXiv) — Dual-branch luminance-guided extreme low-light GS enhancement

- **EnerGS** [arXiv:2604.26238](https://arxiv.org/abs/2604.26238) (Under review) — Energy-based soft geometric guidance from LiDAR/depth priors

- **EnerGS** [arXiv:2604.26238](https://arxiv.org/abs/2604.26238) (Under review) — Energy-based soft geometric guidance from LiDAR/depth priors

- **Energy-GS** [no arXiv ID; verify primary source] (CVPR) — RGB-only pose-free 3DGS joint optimization; no depth/geometry priors required; energy-based formulation - **Ev-GS** [arXiv:2407.11343](https://arxiv.org/abs/2407.11343) (CVPR 202

- **Energy-GS** [no arXiv ID; verify primary source] (CVPR) — RGB-only pose-free 3DGS joint optimization; no depth/geometry priors required; energy-based formulation - **Ev-GS** [arXiv:2407.11343](https://arxiv.org/abs/2407.11343) (CVPR 202

- **Ev-GS** [arXiv:2407.11343](https://arxiv.org/abs/2407.11343) (CVPR 2024) — Event camera-integrated 3DGS for high-speed and HDR scene reconstruction

- **Ev-GS** [arXiv:2407.11343](https://arxiv.org/abs/2407.11343) (CVPR 2024) — Event camera-integrated 3DGS for high-speed and HDR scene reconstruction

- **Faster-GS** [no arXiv ID; verify primary source] (CVPR 2026) — Systematic benchmark separating engineering from algorithmic acceleration - **FastGS** [arXiv:2511.04283](https://arxiv.org/abs/2511.04283) (CVPR) — Multi-view consistency-b

- **Faster-GS** [no arXiv ID; verify primary source] (CVPR 2026) — Systematic benchmark separating engineering from algorithmic acceleration - **FastGS** [arXiv:2511.04283](https://arxiv.org/abs/2511.04283) (CVPR) — Multi-view consistency-b

- **FastGS** [arXiv:2511.04283](https://arxiv.org/abs/2511.04283) (CVPR) — Multi-view consistency-based densification/pruning replacing Gaussian budget; 3.32x training acceleration

- **FastGS** [arXiv:2511.04283](https://arxiv.org/abs/2511.04283) (CVPR) — Multi-view consistency-based densification/pruning replacing Gaussian budget; 3.32x training acceleration

- **FreeFix** [arXiv:2601.20857](https://arxiv.org/abs/2601.20857) (arXiv) — Fine-tuning-free diffusion guidance for extrapolated 3DGS

- **FreeFix** [arXiv:2601.20857](https://arxiv.org/abs/2601.20857) (arXiv) — Fine-tuning-free diffusion guidance for extrapolated 3DGS

- **GaussHDR** [arXiv:2503.10143](https://arxiv.org/abs/2503.10143) (CVPR 2025) — HDR-robust GS with exposure-aware Gaussian decomposition

- **GaussHDR** [arXiv:2503.10143](https://arxiv.org/abs/2503.10143) (CVPR 2025) — HDR-robust GS with exposure-aware Gaussian decomposition

- **GaussianPretrain** [arXiv:2411.12452](https://arxiv.org/abs/2411.12452) (CVPR 2024) — Self-supervised pre-training for Gaussian initialization from multi-view features

- **GaussianPretrain** [arXiv:2411.12452](https://arxiv.org/abs/2411.12452) (CVPR 2024) — Self-supervised pre-training for Gaussian initialization from multi-view features

- **GaussianSR** [arXiv:2406.10111](https://arxiv.org/abs/2406.10111) (CVPR 2024) — 2D super-resolution diffusion prior guiding 3DGS iterative refinement for high-quality rendering

- **GaussianSR** [arXiv:2406.10111](https://arxiv.org/abs/2406.10111) (CVPR 2024) — 2D super-resolution diffusion prior guiding 3DGS iterative refinement for high-quality rendering

- **GEMM-GS** [arXiv:2604.02120](https://arxiv.org/abs/2604.02120) (DAC 2026) — GEMM-compatible blending for Tensor Core utilization, 1.42x speedup

- **GEMM-GS** [arXiv:2604.02120](https://arxiv.org/abs/2604.02120) (DAC 2026) — GEMM-compatible blending for Tensor Core utilization, 1.42x speedup

- **GS-Blur** [arXiv:2410.23658](https://arxiv.org/abs/2410.23658) (CVPR 2025) — Motion blur-aware GS training with blur kernel estimation for sharp reconstruction

- **GS-Blur** [arXiv:2410.23658](https://arxiv.org/abs/2410.23658) (CVPR 2025) — Motion blur-aware GS training with blur kernel estimation for sharp reconstruction

- **GS-PT** [arXiv:2409.04963](https://arxiv.org/abs/2409.04963) (ECCV 2024) — Gaussian Splatting pre-training: self-supervised representation learning for Gaussian initialization

- **GS-PT** [arXiv:2409.04963](https://arxiv.org/abs/2409.04963) (ECCV 2024) — Gaussian Splatting pre-training: self-supervised representation learning for Gaussian initialization

- **HarmoGS** [arXiv:2605.13073](https://arxiv.org/abs/2605.13073) (arXiv 2026) — Conflict-aware gradient harmonization for in-the-wild 3DGS: semantic consistency-guided masking + dual-view gradient rotation

- **HarmoGS** [arXiv:2605.13073](https://arxiv.org/abs/2605.13073) (arXiv 2026) — Conflict-aware gradient harmonization for in-the-wild 3DGS: semantic consistency-guided masking + dual-view gradient rotation

- **Hybrid-Capture Two-View Training** [arXiv:2605.00052](https://arxiv.org/abs/2605.00052) (arXiv) — Two-view-per-step as dominant training lever

- **Hybrid-Capture Two-View Training** [arXiv:2605.00052](https://arxiv.org/abs/2605.00052) (arXiv) — Two-view-per-step as dominant training lever

- **Laplacian Frequency Hierarchies** [arXiv:2609.03334](https://arxiv.org/abs/2609.03334) (Pacific Graphics 2026) — Laplacian frequency hierarchies for efficient 3DGS training; coarse-to-fine frequency-staged, 1.73x speedup

- **Laplacian Frequency Hierarchies** [arXiv:2609.03334](https://arxiv.org/abs/2609.03334) (Pacific Graphics 2026) — Laplacian frequency hierarchies for efficient 3DGS training; coarse-to-fine frequency-staged, 1.73x speedup

- **LE3D** [arXiv:2406.06216](https://arxiv.org/abs/2406.06216) (NeurIPS 2024) — Low-light to HDR linear GS encoding + denoising for Lighting Every Darkness

- **LE3D** [arXiv:2406.06216](https://arxiv.org/abs/2406.06216) (NeurIPS 2024) — Low-light to HDR linear GS encoding + denoising for Lighting Every Darkness

- **Learn2Splat** [arXiv:2605.15760](https://arxiv.org/abs/2605.15760) (arXiv 2026) — Meta-learned optimizer for 3DGS via checkpoint buffer + optimizer rollout; extends optimization horizon without degradation; zero-shot generalization (Gei

- **Learn2Splat** [arXiv:2605.15760](https://arxiv.org/abs/2605.15760) (arXiv 2026) — Meta-learned optimizer for 3DGS via checkpoint buffer + optimizer rollout; extends optimization horizon without degradation; zero-shot generalization (Gei

- **Luminance-GS** [arXiv:2602.18322](https://arxiv.org/abs/2602.18322) (TPAMI 2026) — View-adaptive color/lightness correction for robust NVS

- **Luminance-GS** [arXiv:2602.18322](https://arxiv.org/abs/2602.18322) (TPAMI 2026) — View-adaptive color/lightness correction for robust NVS

- **MarineSTD-GS** [arXiv:2604.23551](https://arxiv.org/abs/2604.23551) (ACM MM 2025) — Spatiotemporal underwater degradation modeling

- **MarineSTD-GS** [arXiv:2604.23551](https://arxiv.org/abs/2604.23551) (ACM MM 2025) — Spatiotemporal underwater degradation modeling

- **MERID-GS** [arXiv:2604.24053](https://arxiv.org/abs/2604.24053) (arXiv) — Retinex-based illumination/reflectance decoupling for low-light

- **MERID-GS** [arXiv:2604.24053](https://arxiv.org/abs/2604.24053) (arXiv) — Retinex-based illumination/reflectance decoupling for low-light

- **NRGS** [arXiv:2604.22439](https://arxiv.org/abs/2604.22439) (arXiv) — Neural regularization for semantic 3DGS

- **NRGS** [arXiv:2604.22439](https://arxiv.org/abs/2604.22439) (arXiv) — Neural regularization for semantic 3DGS

- **PDF-GS** [arXiv:2604.12580](https://arxiv.org/abs/2604.12580) (arXiv) — Progressive distractor filtering for robust 3DGS

- **PDF-GS** [arXiv:2604.12580](https://arxiv.org/abs/2604.12580) (arXiv) — Progressive distractor filtering for robust 3DGS

- **PocketGS** [arXiv:2601.17354](https://arxiv.org/abs/2601.17354) (arXiv) — On-device 3DGS training for mobile with geometry-faithful priors + anisotropic seeding

- **PocketGS** [arXiv:2601.17354](https://arxiv.org/abs/2601.17354) (arXiv) — On-device 3DGS training for mobile with geometry-faithful priors + anisotropic seeding

- **Proxy-GS** [no arXiv ID; verify primary source] (CVPR 2026) — Lightweight proxy model for 2.5x speedup with no accuracy loss - **Prune Wisely** [arXiv:2602.24136](https://arxiv.org/abs/2602.24136) (arXiv 2026) — Adaptive pruning + Diffe

- **Proxy-GS** [no arXiv ID; verify primary source] (CVPR 2026) — Lightweight proxy model for 2.5x speedup with no accuracy loss - **Prune Wisely** [arXiv:2602.24136](https://arxiv.org/abs/2602.24136) (arXiv 2026) — Adaptive pruning + Diffe

- **Prune Wisely** [arXiv:2602.24136](https://arxiv.org/abs/2602.24136) (arXiv 2026) — Adaptive pruning + Difference-of-Gaussian primitives for compact 3DGS; up to 90% Gaussian reduction

- **Prune Wisely** [arXiv:2602.24136](https://arxiv.org/abs/2602.24136) (arXiv 2026) — Adaptive pruning + Difference-of-Gaussian primitives for compact 3DGS; up to 90% Gaussian reduction

- **QuadBox** [arXiv:2605.04844](https://arxiv.org/abs/2605.04844) (arXiv 2026) — Geometry-aware AABB boxes for 1.85x rendering acceleration

- **QuadBox** [arXiv:2605.04844](https://arxiv.org/abs/2605.04844) (arXiv 2026) — Geometry-aware AABB boxes for 1.85x rendering acceleration

- **Rendering-Aware Bayesian 3DGS** [no arXiv ID; verify primary source] (arXiv) — Normal-Inverse-Wishart posterior tracking for Gaussian geometry; native predictive uncertainty for interval calibration + active view selection; ~17x coverag

- **Rendering-Aware Bayesian 3DGS** [no arXiv ID; verify primary source] (arXiv) — Normal-Inverse-Wishart posterior tracking for Gaussian geometry; native predictive uncertainty for interval calibration + active view selection; ~17x coverag

- **ReorgGS** [arXiv:2605.08739](https://arxiv.org/abs/2605.08739) (arXiv 2026) — Equivalent distribution reorganization fixing parameterization degeneration; kNN anisotropic covariance re-estimation

- **ReorgGS** [arXiv:2605.08739](https://arxiv.org/abs/2605.08739) (arXiv 2026) — Equivalent distribution reorganization fixing parameterization degeneration; kNN anisotropic covariance re-estimation

- **RouteBridge** [arXiv:2609.09606](https://arxiv.org/abs/2609.09606) (arXiv 2026) — Reliability-routed bidirectional distillation between NeRF and 3DGS; complementary inductive biases via adaptive routing

- **RouteBridge** [arXiv:2609.09606](https://arxiv.org/abs/2609.09606) (arXiv 2026) — Reliability-routed bidirectional distillation between NeRF and 3DGS; complementary inductive biases via adaptive routing

- **SplatFields** [arXiv:2409.11211](https://arxiv.org/abs/2409.11211) (ECCV 2024) — Implicit neural field regularization on splat features for sparse-view 3D/4D reconstruction

- **SplatFields** [arXiv:2409.11211](https://arxiv.org/abs/2409.11211) (ECCV 2024) — Implicit neural field regularization on splat features for sparse-view 3D/4D reconstruction

- **Structure-Aware Densification** [arXiv:2604.28016](https://arxiv.org/abs/2604.28016) (SIGGRAPH 2026) — Frequency-aware anisotropic split replacing heuristic clone/split

- **Structure-Aware Densification** [arXiv:2604.28016](https://arxiv.org/abs/2604.28016) (SIGGRAPH 2026) — Frequency-aware anisotropic split replacing heuristic clone/split

- **TangoGS** [arXiv:2609.31248](https://arxiv.org/abs/2609.31248) (arXiv 2026) — Compact Gaussian Splatting across scene scales; capture-derived learning allowance from total pixels after discounting re-observing views + training-quality g

- **TangoGS** [arXiv:2609.31248](https://arxiv.org/abs/2609.31248) (arXiv 2026) — Compact Gaussian Splatting across scene scales; capture-derived learning allowance from total pixels after discounting re-observing views + training-quality g

- **TruncGradGS** [arXiv:2609.03534](https://arxiv.org/abs/2609.03534) (Pacific Graphics 2026) — Improved 3DGS via truncated gradient updates; addresses gradient vanishing for robust primitive learning

- **TruncGradGS** [arXiv:2609.03534](https://arxiv.org/abs/2609.03534) (Pacific Graphics 2026) — Improved 3DGS via truncated gradient updates; addresses gradient vanishing for robust primitive learning

- **TWINGS** [arXiv:2605.22069](https://arxiv.org/abs/2605.22069) (CVPR) — Thin Plate Splines warp-aligned initialization for sparse-view 3DGS

- **TWINGS** [arXiv:2605.22069](https://arxiv.org/abs/2605.22069) (CVPR) — Thin Plate Splines warp-aligned initialization for sparse-view 3DGS

- **VkSplat** [arXiv:2605.00219](https://arxiv.org/abs/2605.00219) (Eurographics 2026 Short Papers) — Vulkan-based 3DGS training, 3.3x faster, 33% less VRAM

- **VkSplat** [arXiv:2605.00219](https://arxiv.org/abs/2605.00219) (Eurographics 2026 Short Papers) — Vulkan-based 3DGS training, 3.3x faster, 33% less VRAM

- **WildGaussians** [arXiv:2407.08447](https://arxiv.org/abs/2407.08447) (NeurIPS 2024) — Joint pose + 3DGS optimization from internet photos

- **WildGaussians** [arXiv:2407.08447](https://arxiv.org/abs/2407.08447) (NeurIPS 2024) — Joint pose + 3DGS optimization from internet photos

- **YOGO** [arXiv:2604.21400](https://arxiv.org/abs/2604.21400) (arXiv) — Budget-aware equilibrium for ultra-dense scene control

- **YOGO** [arXiv:2604.21400](https://arxiv.org/abs/2604.21400) (arXiv) — Budget-aware equilibrium for ultra-dense scene control

- **3DReflecNet** [arXiv:2605.10204](https://arxiv.org/abs/2605.10204) (CVPR) — Large-scale dataset for 3D reconstruction from reflection-corrupted images

- **3DReflecNet** [arXiv:2605.10204](https://arxiv.org/abs/2605.10204) (CVPR) — Large-scale dataset for 3D reconstruction from reflection-corrupted images

- **ClearGS** [arXiv:2609.31509](https://arxiv.org/abs/2609.31509) (arXiv 2026) — Reliability-aware Gaussian Splatting from handheld videos; Reliability-aware View Allocation (RVA) with graded supervision weights + Render-Guided In-Video Re

- **ClearGS** [arXiv:2609.31509](https://arxiv.org/abs/2609.31509) (arXiv 2026) — Reliability-aware Gaussian Splatting from handheld videos; Reliability-aware View Allocation (RVA) with graded supervision weights + Render-Guided In-Video Re

- **Dehaze-GaussianImage** [arXiv:2606.16163](https://arxiv.org/abs/2606.16163) (arXiv 2026) — To address these issues, we propose Dehaze-GaussianImage, the first zero-shot framework that introduces 2D Gaussian Splatting (2DGS) into the ima

- **Dehaze-GaussianImage** [arXiv:2606.16163](https://arxiv.org/abs/2606.16163) (arXiv 2026) — To address these issues, we propose Dehaze-GaussianImage, the first zero-shot framework that introduces 2D Gaussian Splatting (2DGS) into the ima

- **DelowlightSplat** [arXiv:2605.26629](https://arxiv.org/abs/2605.26629) (arXiv) — Lowlight Adapter + cost-volume inference predicts clean Gaussians from degraded inputs

- **DelowlightSplat** [arXiv:2605.26629](https://arxiv.org/abs/2605.26629) (arXiv) — Lowlight Adapter + cost-volume inference predicts clean Gaussians from degraded inputs

- **DualPhys-GS** [arXiv:2508.09610](https://arxiv.org/abs/2508.09610) (arXiv 2026) — Dual physically-guided 3DGS for underwater scene reconstruction; joint attenuation-scattering physics + lighting decomposition

- **DualPhys-GS** [arXiv:2508.09610](https://arxiv.org/abs/2508.09610) (arXiv 2026) — Dual physically-guided 3DGS for underwater scene reconstruction; joint attenuation-scattering physics + lighting decomposition

- **Fi-Gaussian** [arXiv:2606.16168](https://arxiv.org/abs/2606.16168) (arXiv 2026) — To address these issues, we propose Fi-Gaussian, a frequency-aware implicit Gaussian splatting network for single image dehazing

- **Fi-Gaussian** [arXiv:2606.16168](https://arxiv.org/abs/2606.16168) (arXiv 2026) — To address these issues, we propose Fi-Gaussian, a frequency-aware implicit Gaussian splatting network for single image dehazing

- **FreDeGS** [no arXiv ID; verify primary source] (Springer 2026) — Frequency-guided 3DGS for scene deblurring; frequency-aware decomposition for motion-blurred input reconstruction - **FujinSplat** [arXiv:2609.06017](https://arxiv.org/abs

- **FreDeGS** [no arXiv ID; verify primary source] (Springer 2026) — Frequency-guided 3DGS for scene deblurring; frequency-aware decomposition for motion-blurred input reconstruction - **FujinSplat** [arXiv:2609.06017](https://arxiv.org/abs

- **FujinSplat** [arXiv:2609.06017](https://arxiv.org/abs/2609.06017) (arXiv 2026) — RAW-domain Gaussian splatting seeing through smoke; separates participating medium and ISP in RAW domain

- **FujinSplat** [arXiv:2609.06017](https://arxiv.org/abs/2609.06017) (arXiv 2026) — RAW-domain Gaussian splatting seeing through smoke; separates participating medium and ISP in RAW domain

- **GlowGS** [arXiv:2605.23602](https://arxiv.org/abs/2605.23602) (CVPR Findings) — Diffusion + VFM semantic features for degraded-image 3DGS reconstruction

- **GlowGS** [arXiv:2605.23602](https://arxiv.org/abs/2605.23602) (CVPR Findings) — Diffusion + VFM semantic features for degraded-image 3DGS reconstruction

- **JADE-GS** [arXiv:2607.14990](https://arxiv.org/abs/2607.14990) (arXiv 2026) — Formulates the combination of these priors as spatial evidence allocation

- **JADE-GS** [arXiv:2607.14990](https://arxiv.org/abs/2607.14990) (arXiv 2026) — Formulates the combination of these priors as spatial evidence allocation

- **Mind the Gap** [no arXiv ID; verify primary source] (arXiv) — Reveals 3-12 dB interpolation-extrapolation gap in standard 3DGS evaluation; proposes fair match count protocol + spatial preservation benchmark toolkit (16 scenes) - **MLLM-

- **Mind the Gap** [no arXiv ID; verify primary source] (arXiv) — Reveals 3-12 dB interpolation-extrapolation gap in standard 3DGS evaluation; proposes fair match count protocol + spatial preservation benchmark toolkit (16 scenes) - **MLLM-

- **MLLM-GS** [arXiv:2610.39936](https://arxiv.org/abs/2610.39936) (arXiv 2026) — MLLM prior-guided framework for fast NVS in extreme environments; a frozen MLLM with lightweight correction turns degraded views into pseudo-GT supervision, w

- **MLLM-GS** [arXiv:2610.39936](https://arxiv.org/abs/2610.39936) (arXiv 2026) — MLLM prior-guided framework for fast NVS in extreme environments; a frozen MLLM with lightweight correction turns degraded views into pseudo-GT supervision, w

- **NIRRGB-GS** [no arXiv ID; verify primary source] (AISY 2026) — Near-infrared assisted low-light scene reconstruction and enhancement via multi-modal NIR+RGB Gaussian Splatting - **NOVA-GS** [arXiv:2609.12682](https://arxiv.org/abs/2609.

- **NIRRGB-GS** [no arXiv ID; verify primary source] (AISY 2026) — Near-infrared assisted low-light scene reconstruction and enhancement via multi-modal NIR+RGB Gaussian Splatting - **NOVA-GS** [arXiv:2609.12682](https://arxiv.org/abs/2609.

- **NOVA-GS** [arXiv:2609.12682](https://arxiv.org/abs/2609.12682) (arXiv 2026) — Noise-aware view-consistent 3DGS for low-light novel view synthesis; handles severe sensor noise and low SNR

- **NOVA-GS** [arXiv:2609.12682](https://arxiv.org/abs/2609.12682) (arXiv 2026) — Noise-aware view-consistent 3DGS for low-light novel view synthesis; handles severe sensor noise and low SNR

- **Per-View Distractor Filtering** [arXiv:2608.26951](https://arxiv.org/abs/2608.26951) (arXiv 2026) — Training-free distractor filtering in feed-forward 3DGS; excludes per-view Gaussians and renders with remaining representation to identi

- **Per-View Distractor Filtering** [arXiv:2608.26951](https://arxiv.org/abs/2608.26951) (arXiv 2026) — Training-free distractor filtering in feed-forward 3DGS; excludes per-view Gaussians and renders with remaining representation to identi

- **PriSplat** [no arXiv ID; verify primary source] (ECCV 2026) — Propagating reliable multi-view information for distractor-free 3DGS; 3D-aware inpainting engine from view synthesis prior; geometry-aware support view selection; mask-aware 

- **PriSplat** [no arXiv ID; verify primary source] (ECCV 2026) — Propagating reliable multi-view information for distractor-free 3DGS; 3D-aware inpainting engine from view synthesis prior; geometry-aware support view selection; mask-aware 

- **PUGS** [arXiv:2610.17936](https://arxiv.org/abs/2610.17936) (arXiv 2026) — Physics-guided uncertainty modeling for underwater dynamic 3DGS; stochastic dynamics replace deterministic deformation under scattering, refraction, and marine s

- **PUGS** [arXiv:2610.17936](https://arxiv.org/abs/2610.17936) (arXiv 2026) — Physics-guided uncertainty modeling for underwater dynamic 3DGS; stochastic dynamics replace deterministic deformation under scattering, refraction, and marine s

- **SharpSplat** [no arXiv ID; verify primary source] (arXiv) — SAM3-based building edge extraction + rendered gradient-edge alignment; sharp building boundary reconstruction in UAV 3DGS scenes - **SSA-3DGS** [no arXiv ID; verify primary so

- **SharpSplat** [no arXiv ID; verify primary source] (arXiv) — SAM3-based building edge extraction + rendered gradient-edge alignment; sharp building boundary reconstruction in UAV 3DGS scenes - **SSA-3DGS** [no arXiv ID; verify primary so

- **SSA-3DGS** [no arXiv ID; verify primary source] (arXiv) — Unsupervised screen-space artifact removal for 3DGS; joint optimization of 3D scene + learnable 2D overlay; cross-view geometric consistency decouples artifacts from geometry; up

- **SSA-3DGS** [no arXiv ID; verify primary source] (arXiv) — Unsupervised screen-space artifact removal for 3DGS; joint optimization of 3D scene + learnable 2D overlay; cross-view geometric consistency decouples artifacts from geometry; up

- **TDg** [arXiv:2605.30328](https://arxiv.org/abs/2605.30328) (ISPRS) — Thermal-to-Depth GS using only thermal images for depth estimation

- **TDg** [arXiv:2605.30328](https://arxiv.org/abs/2605.30328) (ISPRS) — Thermal-to-Depth GS using only thermal images for depth estimation

- **Tri-DehazeGS** [arXiv:2609.11223](https://arxiv.org/abs/2609.11223) (arXiv 2026) — Scene-medium decoupled 3DGS with transmittance-aware optimization for hazy multi-view reconstruction

- **Tri-DehazeGS** [arXiv:2609.11223](https://arxiv.org/abs/2609.11223) (arXiv 2026) — Scene-medium decoupled 3DGS with transmittance-aware optimization for hazy multi-view reconstruction

- **Underwater360** [arXiv:2605.26447](https://arxiv.org/abs/2605.26447) (arXiv) — Omnidirectional GS + physics-based appearance-medium decoupling for underwater panoramic scenes

- **Underwater360** [arXiv:2605.26447](https://arxiv.org/abs/2605.26447) (arXiv) — Omnidirectional GS + physics-based appearance-medium decoupling for underwater panoramic scenes

- **View-Structured Conformal Prediction** [arXiv:2609.10307](https://arxiv.org/abs/2609.10307) (arXiv 2026) — View-structured conformal prediction for 3DGS; certifies rendered view meets prediction coverage via statistical guarantee

- **View-Structured Conformal Prediction** [arXiv:2609.10307](https://arxiv.org/abs/2609.10307) (arXiv 2026) — View-structured conformal prediction for 3DGS; certifies rendered view meets prediction coverage via statistical guarantee

- **WilLaGS** [arXiv:2608.28240](https://arxiv.org/abs/2608.28240) (arXiv 2026) — To address these limitations, we propose textbfWilLaGS, a unified framework for robust 3D scene reconstruction and generative appearance synthesis under uncon

- **WilLaGS** [arXiv:2608.28240](https://arxiv.org/abs/2608.28240) (arXiv 2026) — To address these limitations, we propose textbfWilLaGS, a unified framework for robust 3D scene reconstruction and generative appearance synthesis under uncon

- **3DThinkVLA** [arXiv:2606.04436](https://arxiv.org/abs/2606.04436) (arXiv) — Endowing VLA models with latent 3D priors via 3D-thinking-guided co-training; SOTA on LIBERO

- **3DThinkVLA** [arXiv:2606.04436](https://arxiv.org/abs/2606.04436) (arXiv) — Endowing VLA models with latent 3D priors via 3D-thinking-guided co-training; SOTA on LIBERO

- **AGILE** [arXiv:2602.04672](https://arxiv.org/abs/2602.04672) (SIGGRAPH) — Agentic generation for hand-object interaction reconstruction from video; VLM guides generative model

- **AGILE** [arXiv:2602.04672](https://arxiv.org/abs/2602.04672) (SIGGRAPH) — Agentic generation for hand-object interaction reconstruction from video; VLM guides generative model

- **ArtiTwinSplat** [no arXiv ID; verify primary source] (arXiv) — Interactable Digital Twin Reconstruction via Gaussian Splatting from RGB-D videos; enables interaction with reconstructed digital twins; agent-system integration for articul

- **ArtiTwinSplat** [no arXiv ID; verify primary source] (arXiv) — Interactable Digital Twin Reconstruction via Gaussian Splatting from RGB-D videos; enables interaction with reconstructed digital twins; agent-system integration for articul

- **DISCOVERSE** [no arXiv ID; verify primary source] (arXiv) — 3DGS + MuJoCo unified Real2Sim2Real robot learning framework; scene-level + object-level digital twin - **Dynamic Mesh-Gaussian** [arXiv:2606.00444](https://arxiv.org/abs/2606.

- **DISCOVERSE** [no arXiv ID; verify primary source] (arXiv) — 3DGS + MuJoCo unified Real2Sim2Real robot learning framework; scene-level + object-level digital twin - **Dynamic Mesh-Gaussian** [arXiv:2606.00444](https://arxiv.org/abs/2606.

- **Dynamic Mesh-Gaussian** [arXiv:2606.00444](https://arxiv.org/abs/2606.00444) (arXiv 2026) — Dual-representation framework: fixed-topology mesh for physics + Gaussian splatting for rendering; 4.65x speedup over varying-topology baseline

- **Dynamic Mesh-Gaussian** [arXiv:2606.00444](https://arxiv.org/abs/2606.00444) (arXiv 2026) — Dual-representation framework: fixed-topology mesh for physics + Gaussian splatting for rendering; 4.65x speedup over varying-topology baseline

- **FLUIDSPLAT** [arXiv:2605.18866](https://arxiv.org/abs/2605.18866) (arXiv 2026) — Physical flow field reconstruction via anisotropic GS partition-of-unity; proven Sobolev approximation rate

- **FLUIDSPLAT** [arXiv:2605.18866](https://arxiv.org/abs/2605.18866) (arXiv 2026) — Physical flow field reconstruction via anisotropic GS partition-of-unity; proven Sobolev approximation rate

- **LaGSplat** [arXiv:2608.16324](https://arxiv.org/abs/2608.16324) (arXiv 2026) — Framework that infers interactive, physics-governed dynamics from one or a few monocular videos

- **LaGSplat** [arXiv:2608.16324](https://arxiv.org/abs/2608.16324) (arXiv 2026) — Framework that infers interactive, physics-governed dynamics from one or a few monocular videos

- **MonoPhysics** [arXiv:2605.30320](https://arxiv.org/abs/2605.30320) (arXiv) — Monocular inverse physics using differentiable MPM + 3DGS; three visual-physical bridges

- **MonoPhysics** [arXiv:2605.30320](https://arxiv.org/abs/2605.30320) (arXiv) — Monocular inverse physics using differentiable MPM + 3DGS; three visual-physical bridges

- **PhyGenHOI** [arXiv:2605.30268](https://arxiv.org/abs/2605.30268) (arXiv) — Physically-aware 4D HOI generation; MDM + MPM + 3DGS unified; windowed attraction + re-simulation

- **PhyGenHOI** [arXiv:2605.30268](https://arxiv.org/abs/2605.30268) (arXiv) — Physically-aware 4D HOI generation; MDM + MPM + 3DGS unified; windowed attraction + re-simulation

- **SceneAgent** [arXiv:2610.34745](https://arxiv.org/abs/2610.34745) (arXiv 2026) — Agentic pipeline converting 3D/4D captures (3DGS, photogrammetry, LiDAR, 4D video) into interactive simulation environments through simulator-based visual 

- **SceneAgent** [arXiv:2610.34745](https://arxiv.org/abs/2610.34745) (arXiv 2026) — Agentic pipeline converting 3D/4D captures (3DGS, photogrammetry, LiDAR, 4D video) into interactive simulation environments through simulator-based visual 

- **SimuScene** [arXiv:2606.03994](https://arxiv.org/abs/2606.03994) (arXiv) — Physics-in-the-loop compositional 3D scene reconstruction from single image; physics engine as diagnostic

- **SimuScene** [arXiv:2606.03994](https://arxiv.org/abs/2606.03994) (arXiv) — Physics-in-the-loop compositional 3D scene reconstruction from single image; physics engine as diagnostic

- **WildFireGS** [arXiv:2608.11100](https://arxiv.org/abs/2608.11100) (arXiv 2026) — Particle-based combustion model that operates natively on Gaussian representations, simulating ignition, heat transfer, combustion, and flame propagation a

- **WildFireGS** [arXiv:2608.11100](https://arxiv.org/abs/2608.11100) (arXiv 2026) — Particle-based combustion model that operates natively on Gaussian representations, simulating ignition, heat transfer, combustion, and flame propagation a

- **3DTV** [arXiv:2604.11211](https://arxiv.org/abs/2604.11211) (arXiv) — Feedforward 3-camera interpolation at 40 FPS