Research map / Point clouds / Segmentation and detection
Fast Point R-CNN
Yilun Chen, Shu Liu, Xiaoyong Shen, Jiaya Jia
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 3D convolutions in lower layers (ablation) | VoxelRPN baseline · KITTI val | 3D AP0.7 moderate: 73.8 -> 74.7 | Improved |
| add higher resolution voxel input (ablation) | VoxelRPN · KITTI val | 3D AP0.7 moderate: 74.7 -> 75.34 | Improved |
| add MIXUP augmentation (ablation) | VoxelRPN · KITTI val | 3D AP0.7 moderate: 75.34 -> 75.82 | Improved |
| use 4 anchor angles instead of 2 (ablation) | VoxelRPN · KITTI val | 3D AP0.7 moderate: 75.82 -> 76.64 | Improved |
| add RefinerNet second stage (ablation) | VoxelRPN · KITTI val | 3D AP0.7 moderate: 76.64 -> 79.0 | Improved |
| add RefinerNet second stage (ablation) | VoxelRPN · KITTI val | 3D AP0.8 moderate: 42.6 -> 52.95 | Improved |
| add RefinerNet second stage for distant objects (30-50 m) (ablation) | VoxelRPN · KITTI val | 3D AP0.7 moderate: 51.99 -> 58.41 | Improved |
| add RefinerNet second stage for BEV at loose IoU (ablation) | VoxelRPN · KITTI val | BEV AP0.7 moderate (0-30 m): 90.22 -> 90.25 | No clear effect |
| fuse coordinate and convolution features with attention instead of concatenation (ablation) | RefinerNet · KITTI val | 3D AP0.7 moderate: 78.38 -> 79.00 | Improved |
| add convolution feature to coordinate feature (concatenation) (ablation) | RefinerNet · KITTI val | 3D AP0.7 moderate: 77.82 -> 78.38 | Improved |
| use canonized corner loss instead of naive 7-parameter regression (ablation) | RefinerNet · KITTI val | 3D AP0.7 moderate: 78.45 -> 79 | Improved |
| use RefinerNet instead of rotated RoI Align refinement (ablation) | Fast Point R-CNN · KITTI val | 3D AP0.7 moderate: 77.39 -> 79.00 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.
- PointPillars — PointPillars: Fast Encoders for Object Detection From Point Clouds (CVPR 2019) · better on 2 of 6 shared results
- PointRCNN — PointRCNN: 3D Object Proposal Generation and Detection From Point Cloud (CVPR 2019) · better on 1 of 6 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
- VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection (CVPR 2018 · similarity 0.92)
- LiDAR R-CNN: An Efficient and Universal 3D Object Detector (CVPR 2021 · similarity 0.92)
- Graph R-CNN: Towards Accurate 3D Object Detection with Semantic-Decorated Local Graph (ECCV 2022 · similarity 0.91)
- HVPR: Hybrid Voxel-Point Representation for Single-Stage 3D Object Detection (CVPR 2021 · similarity 0.91)
- JPV-Net: Joint Point-Voxel Representations for Accurate 3D Object Detection (AAAI 2022 · similarity 0.91)
- Pyramid R-CNN: Towards Better Performance and Adaptability for 3D Object Detection (ICCV 2021 · similarity 0.91)
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.
