Research map / 3D vision / 3D generation
GaussianFormer: Scene as Gaussians for Vision-Based 3D Semantic Occupancy Prediction
Yuanhui Huang, Wenzhao Zheng, Yunpeng Zhang, Jie Zhou, Jiwen Lu
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 |
|---|---|---|---|
| use 3D Gaussian representation instead of dense grids for occupancy | GaussianFormer · nuScenes efficiency | memory (M): 26500 M (OctreeOcc) -> 6229 M | Improved |
| GaussianFormer vs best dense method on accuracy | GaussianFormer · nuScenes val | SSC mIoU: 20.30 -> 19.10 | Hurt |
| GaussianFormer vs planar representation BEVFormer | GaussianFormer · nuScenes val | SSC mIoU: 16.75 -> 19.10 | Improved |
| GaussianFormer on monocular KITTI-360 vs Symphonies | GaussianFormer · SSCBench-KITTI-360 val | SSC mIoU: 18.58 -> 12.92 | Hurt |
| add 3D sparse convolution self-encoding (ablation) | GaussianFormer · nuScenes val (51200 Gaussians) | mIoU: 15.93 -> 16.41 | Improved |
| add deep supervision on each refinement module (ablation) | GaussianFormer · nuScenes val (51200 Gaussians) | mIoU: 16.36 -> 16.41 | Improved |
| refine properties by direct substitution instead of residual mean refinement (ablation) | GaussianFormer · nuScenes val (51200 Gaussians) | mIoU: collapses (no result) | Hurt |
| apply residual refinement to all properties except semantics (ablation) | GaussianFormer · nuScenes val (51200 Gaussians) | mIoU: 16.41 -> 16.24 | Hurt |
| increase number of Gaussians from 25600 to 144000 (ablation) | GaussianFormer · nuScenes val | mIoU: 16.00 -> 19.10 | Improved |
| add photometric reconstruction loss (ablation) | GaussianFormer · nuScenes val (51200 Gaussians) | mIoU: 16.41 -> 16.24 | No clear effect |
| add split and prune strategy from offline 3D-GS (ablation) | GaussianFormer · nuScenes val (51200 Gaussians) | mIoU: 16.41 -> 16.50 | No clear effect |
| initialize Gaussians from predicted pseudo point cloud instead of learnable (ablation) | GaussianFormer · nuScenes val (51200 Gaussians) | mIoU: 16.41 -> 18.99 | Improved |
Later papers that compared with this method
8 later accepted papers report GaussianFormer as a baseline in their results tables; 8 of them report a better number on most of the results they share with it.
- Generalizing Visual Geometry Priors to Sparse Gaussian Occupancy Prediction (CVPR 2026 · better on 2 of 2 shared results)
- SplatSSC: Decoupled Depth-Guided Gaussian Splatting for Semantic Scene Completion (AAAI 2026 · better on 2 of 2 shared results)
- Sparse-Aware Vector Quantization for Bandwidth-Efficient Collaborative 3D Semantic Occupancy Prediction (ECCV 2026 · better on 2 of 2 shared results)
- DLWM: Dual Latent World Models enable Holistic Gaussian-centric Pre-training in Autonomous Driving (CVPR 2026 · better on 1 of 1 shared result)
- AdaSFormer: Adaptive Serialized Transformers for Monocular Semantic Scene Completion from Indoor Environments (CVPR 2026 · better on 1 of 1 shared result)
- EmbodiedOcc: Embodied 3D Occupancy Prediction for Vision-based Online Scene Understanding (ICCV 2025 · better on 4 of 4 shared results)
- GaussianFormer-2: Probabilistic Gaussian Superposition for Efficient 3D Occupancy Prediction (CVPR 2025 · better on 2 of 2 shared results)
- GaussRender: Learning 3D Occupancy with Gaussian Rendering (ICCV 2025 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 4 on most shared results.
- OctreeOcc — OctreeOcc: Efficient and Multi-Granularity Occupancy Prediction Using Octree Queries (NeurIPS 2024) · better on 2 of 2 shared results
- OccFormer — OccFormer: Dual-path Transformer for Vision-based 3D Semantic Occupancy Prediction (ICCV 2023) · better on 1 of 2 shared results
- SurroundOcc — SurroundOcc: Multi-camera 3D Occupancy Prediction for Autonomous Driving (ICCV 2023) · better on 0 of 2 shared results
- MonoScene — MonoScene: Monocular 3D Semantic Scene Completion (CVPR 2022) · better on 1 of 1 shared result
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
- GaussianFormer-2: Probabilistic Gaussian Superposition for Efficient 3D Occupancy Prediction (CVPR 2025 · similarity 0.96)
- GaussTR: Foundation Model-Aligned Gaussian Transformer for Self-Supervised 3D Spatial Understanding (CVPR 2025 · similarity 0.93)
- ODG: Occupancy Prediction Using Dual Gaussians (NeurIPS 2025 · similarity 0.93)
- GaussianWorld: Gaussian World Model for Streaming 3D Occupancy Prediction (CVPR 2025 · similarity 0.91)
- VoxelSplat: Dynamic Gaussian Splatting as an Effective Loss for Occupancy and Flow Prediction (CVPR 2025 · similarity 0.91)
- QuadricFormer: Scene as Superquadrics for 3D Semantic Occupancy Prediction (NeurIPS 2025 · similarity 0.90)
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.
