Research map / Detection and segmentation / 3D detection and driving
FSD V2: Improving Fully Sparse 3D Object Detection With Virtual Voxels
Lue Fan, Wang Feng, Naiyan Wang, Zhaoxiang Zhang
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 virtual voxel mixer on real voxels (ablation) | FSDv2-Baseline · nuScenes val | mAP: 60.9 -> 62.2 | Improved |
| add virtual voxelization (virtual voxels from voted centers) (ablation) | FSDv2 + Mixer · nuScenes val | mAP: 62.2 -> 62.7 | Improved |
| use weighted centroids for virtual voxel assignment instead of geometric centers (ablation) | FSDv2 · Argoverse 2 val | mAP: 34.3 -> 36.7 | Improved |
| use weighted centroids for assignment instead of geometric centers (ablation) | FSDv2 · Argoverse 2 val | Bollard AP: 34.1 -> 51.4 | Improved |
| add virtual voxel encoder (VVE) (ablation) | FSDv2 · nuScenes val | mAP: 62.1 -> 62.7 | Improved |
| add real voxels as mixer input (ablation) | FSDv2 · nuScenes val | mAP: 60.6 -> 62.0 | Improved |
| use multi-scale real voxels as mixer input (ablation) | FSDv2 · nuScenes val | mAP: 62.0 -> 62.7 | Improved |
| scale center voting offsets from 0 to 1 (full virtual voxel effect) (ablation) | FSDv2 · nuScenes val | Barrier AP: 66.8 -> 72.7 | Improved |
| treat only nearest virtual voxel as positive instead of all in box (ablation) | FSDv2 · Argoverse 2 val | mAP: 36.2 -> 35.0 | Hurt |
| enlarge GT boxes by 0.5m for assignment (ablation) | FSDv2 w/o VA · Argoverse 2 val | mAP: 34.3 -> 33.5 | Hurt |
| use weighted centroid with alpha=0.0 (ablation) | FSDv2 · Argoverse 2 val | mAP: 36.2 -> 36.9 | Improved |
| vary virtual voxel size from (0.4,0.4,0.4) to (0.3,0.3,0.4) (ablation) | FSDv2 · nuScenes val | mAP: 62.0 -> 61.7 | No clear effect |
Later papers that compared with this method
2 later accepted papers report FSDv2 as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- SAFDNet: A Simple and Effective Network for Fully Sparse 3D Object Detection (CVPR 2024 · better on 3 of 3 shared results)
- Towards Flexible 3D Perception: Object-Centric Occupancy Completion Augments 3D Object Detection (NeurIPS 2024 · better on 1 of 1 shared result)
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.
- VoxelNeXt — VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking (CVPR 2023) · better on 7 of 7 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
- Super Sparse 3D Object Detection (TPAMI 2023 · similarity 0.92)
- Fully Sparse 3D Object Detection (NeurIPS 2022 · similarity 0.92)
- VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking (CVPR 2023 · similarity 0.92)
- MsSVT: Mixed-scale Sparse Voxel Transformer for 3D Object Detection on Point Clouds (NeurIPS 2022 · similarity 0.91)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 2024 · similarity 0.90)
- FSHNet: Fully Sparse Hybrid Network for 3D Object Detection (CVPR 2025 · similarity 0.88)
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.
