Research map / Detection and segmentation / 3D detection and driving
Transformation-Equivariant 3D Object Detection for Autonomous Driving
Hai Wu, Chenglu Wen, Wei Li, Xin Li, Ruigang Yang, Cheng Wang
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 transformation-equivariant TED-S instead of Voxel-RCNN | Voxel-RCNN · KITTI val | Car 3D AP (R40), Moderate: 85.29 -> 87.91 | Improved |
| Use transformation-equivariant TED-M instead of multi-modal SFD | multi-modal detector · KITTI val | Car 3D AP (R40), Moderate: 88.27 -> 88.94 | Improved |
| Use TED-M instead of SFD | multi-modal detector · KITTI test | Car 3D AP (R40), Moderate: 84.76 -> 85.28 | Improved |
| Use TED-S instead of PV-RCNN++ | LiDAR-only detector · Waymo val | Vehicle mAP (L2): 70.26 -> 70.50 | Improved |
| Add DA-Aug (ablation) | Voxel-RCNN · KITTI val | Car 3D AP: 85.29 -> 85.72 | Improved |
| Replace SpConv with TeSpConv and add TeBEV pooling (ablation) | Voxel-RCNN + DA-Aug · KITTI val | Car 3D AP: 85.72 -> 86.33 | Improved |
| Add TiVoxel pooling (ablation) | TeSpConv + DA-Aug + TeBEV pooling · KITTI val | Car 3D AP: 86.33 -> 87.91 | Improved |
| Use 1 rotation instead of 3 (ablation) | TED-S · KITTI val | Car 3D AP: 87.91 -> 85.73 | Hurt |
| Use 2 rotations instead of 3 (ablation) | TED-S · KITTI val | Car 3D AP: 87.91 -> 87.13 | Hurt |
| Increase rotation count from 3 to 4 (ablation) | TED-S · KITTI val | Car 3D AP: 87.91 -> 87.95 | No clear effect |
| Increase rotation count from 3 to 4 (ablation) | TED-M · KITTI val | Car 3D AP: 88.94 -> 88.89 | Hurt |
| Replace single-pass detection with TTA (ablation) | Voxel-RCNN with DA-Aug · KITTI val | Car 3D AP, Moderate: 85.72 -> 86.65 | Improved |
Closest papers by meaning
- Rotationally Equivariant 3D Object Detection (CVPR 2022 · similarity 0.89)
- Equivariant Spatio-Temporal Self-Supervision for LiDAR Object Detection (ECCV 2024 · similarity 0.87)
- DEVIANT: Depth EquiVarIAnt NeTwork for Monocular 3D Object Detection (ECCV 2022 · similarity 0.86)
- Viewpoint Equivariance for Multi-View 3D Object Detection (CVPR 2023 · similarity 0.86)
- Equivariant Point Network for 3D Point Cloud Analysis (CVPR 2021 · similarity 0.86)
- R2Det: Exploring Relaxed Rotation Equivariance in 2D Object Detection (ICLR 2025 · similarity 0.86)
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.
