Research map / 3D vision / Gaussian splatting
GS-Occ3D: Scaling Vision-only Occupancy Reconstruction with Gaussian Splatting
Baijun Ye, Minghui Qin, Saining Zhang, Moonjun Gong, Shaoting Zhu, Hao Zhao, Hang Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add octree-based Gaussian surfel with ground Gaussians over GVKF | GS-Occ3D · Waymo Static-32 | Chamfer Distance: 0.82 -> 0.56 | Improved |
| use 5 cameras instead of 3 as input (ablation) | GS-Occ3D · Waymo Static-32 | CD-Pcd: 0.66 -> 0.56 | Improved |
| use 5 cameras instead of 3 as input (ablation) | GVKF · Waymo Static-32 | CD-Pcd: 0.87 -> 0.82 | Hurt |
| use 5 cameras instead of 3 as input (ablation) | PGSR · Waymo Static-32 | CD-Pcd: 2.90 -> 3.63 | Hurt |
| use 5 cameras instead of 3 as input (ablation) | 2DGS · Waymo Static-32 | CD-Pcd: 1.23 -> 1.25 | No clear effect |
| use point cloud directly instead of converting Gaussians to mesh (ablation) | GVKF · Waymo Static-32 (5Cam) | Chamfer Distance: 1.22 -> 0.82 | Improved |
| add separate ground Gaussians (ablation) | GS-based reconstruction · Waymo Static-32 | geometry (qualitative) | Improved |
| train occupancy model on vision-only labels instead of Occ3D LiDAR labels, evaluate zero-shot on nuScenes | CVT-Occ · Occ3D-nuScenes val | IoU: 31.4 -> 33.4 | Improved |
| train occupancy model on vision-only labels instead of Occ3D LiDAR labels, evaluate on Occ3D-Waymo val | CVT-Occ · Occ3D-Val Waymo | IoU: 57.4 -> 44.7 | Hurt |
| train occupancy model on vision-only labels instead of Occ3D labels, evaluate on Ours-Val | CVT-Occ · Ours-Val Waymo | IoU: 41.1 -> 46.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GVKF — GVKF: Gaussian Voxel Kernel Functions for Highly Efficient Surface Reconstruction in Open Scenes (NeurIPS 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ODG: Occupancy Prediction Using Dual Gaussians (NeurIPS 2025 · similarity 0.91)
- CSV-Occ: Fusing Multi-frame Alignment for Occupancy Prediction with Temporal Cross State Space Model and Central Voting Mechanism (ICML 2025 · similarity 0.90)
- SDGOCC: Semantic and Depth-Guided Bird's-Eye View Transformation for 3D Multimodal Occupancy Prediction (CVPR 2025 · similarity 0.90)
- Spatiotemporal Decoupling for Efficient Vision-Based Occupancy Forecasting (CVPR 2025 · similarity 0.90)
- Achieving Speed-Accuracy Balance in Vision-based 3D Occupancy Prediction via Geometric-Semantic Disentanglement (AAAI 2025 · similarity 0.89)
- COTR: Compact Occupancy TRansformer for Vision-based 3D Occupancy Prediction (CVPR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
