Research map / Point clouds / Segmentation and detection
STD: Sparse-to-Dense 3D Object Detector for Point Cloud
Zetong Yang, Yanan Sun, 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 the 3D IoU branch (cls-score x predicted IoU for NMS) instead of plain NMS (ablation) | STD · KITTI val Car | AP 3D moderate (%): 78.7 -> 79.8 | Improved |
| add 3D IoU branch instead of Soft-NMS (ablation) | STD · KITTI val Car | AP 3D moderate (%): 79.0 -> 79.8 | Improved |
| multiply classification score with predicted IoU instead of using predicted 3D IoU alone for NMS (ablation) | STD · KITTI val Car | AP 3D moderate (%): 79.1 -> 79.8 | Improved |
| use oracle IoU instead of classification score for NMS sorting (ablation) | STD · KITTI val Car | AP 3D moderate (%): 78.7 -> 90.9 | Improved |
| add semantic segmentation features to proposal features (ablation) | STD · KITTI val Car | AP 3D moderate (%): 31.1 -> 67.6 | Improved |
| add canonical coordinates to proposal features (ablation) | STD · KITTI val Car | AP 3D moderate (%): 67.6 -> 78.7 | Improved |
| use spherical anchors instead of single-orientation cuboid anchors (ablation) | STD proposal generation · KITTI val Car | recall IoU=0.7: 74.2 -> 76.8 | Improved |
| use spherical anchors instead of 2x cuboid anchors (ablation) | STD proposal generation · KITTI val Car | recall IoU=0.7: 75.7 -> 76.8 | Improved |
| replace PointNet head with PointsPool + 2FC (ablation) | STD box head · KITTI val Car | AP 3D moderate (%): 77.1 -> 78.7 | Improved |
| replace PointNet++ head with PointsPool + 2FC (ablation) | STD box head · KITTI val Car | AP 3D moderate (%): 79.1 -> 78.7 | Mixed |
| merge batch norm into conv layers and split first SA input into 32 x 512 for parallel computation (ablation) | STD · KITTI val Car | inference time: 25ms and 50ms speedup | No clear effect |
| add spherical anchors proposal generation | STD · KITTI val Car | recall IoU=0.5, 50 proposals: 91.0 -> 96.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- PointRCNN — PointRCNN: 3D Object Proposal Generation and Detection From Point Cloud (CVPR 2019) · better on 4 of 4 shared results
- PointPillars — PointPillars: Fast Encoders for Object Detection From Point Clouds (CVPR 2019) · better on 3 of 3 shared results
- VoxelNet — VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection (CVPR 2018) · 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
- Fast Point R-CNN (ICCV 2019 · similarity 0.90)
- Sparse2Dense: Learning to Densify 3D Features for 3D Object Detection (NeurIPS 2022 · similarity 0.90)
- Generative Sparse Detection Networks for 3D Single-shot Object Detection (ECCV 2020 · similarity 0.90)
- Frustum PointNets for 3D Object Detection From RGB-D Data (CVPR 2018 · similarity 0.89)
- PointRCNN: 3D Object Proposal Generation and Detection From Point Cloud (CVPR 2019 · similarity 0.89)
- Structure Aware Single-Stage 3D Object Detection From Point Cloud (CVPR 2020 · 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.
