Research map / Point clouds / Segmentation and detection
Embracing Single Stride 3D Object Detector With Sparse Transformer
Lue Fan, Ziqi Pang, Tianyuan Zhang, Yu-Xiong Wang, Hang Zhao, Feng Wang, Naiyan Wang, Zhaoxiang Zhang
Check 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 |
|---|---|---|---|
| Reduce backbone strides from {1, 2, 4, 8} to {1, 2, 2, 2} (ablation) | PointPillars on Waymo Open Dataset · Waymo Open Dataset validation | vehicle AP: 63.66 -> 66.03 | Improved |
| Use single stride {1, 1, 1, 1} rather than {1, 2, 2, 2} (ablation) | PointPillars on Waymo Open Dataset · Waymo Open Dataset validation | vehicle AP: 66.03 -> 64.69 | Hurt |
| Add dilation 2 in the last two stages (ablation) | Single-stride PointPillars · Waymo Open Dataset validation | vehicle AP: 64.69 -> 66.26 | Improved |
| Add dilation 2 in the last two stages (ablation) | Single-stride PointPillars · Waymo Open Dataset validation | pedestrian AP: 64.32 -> 63.51 | Hurt |
| Use 5 × 5 kernels in the last two stages (ablation) | Single-stride PointPillars · Waymo Open Dataset validation | vehicle AP: 64.69 -> 66.42 | Improved |
| Use 5 × 5 kernels in the last two stages (ablation) | Single-stride PointPillars · Waymo Open Dataset validation | latency: 185ms -> 340ms | Hurt |
| Replace standard convolutions with submanifold sparse convolutions (ablation) | PointPillars-SS · Waymo Open Dataset, 20% training data | vehicle 3D AP: 64.69 -> 51.57 | Hurt |
| Increase sparse-convolution kernel size from 3 × 3 to 5 × 5 (ablation) | SparsePillars-SS · Waymo Open Dataset, 20% training data | vehicle 3D AP: 51.57 -> 55.40 | Improved |
| Increase sparse-convolution kernel size from 3 × 3 to 7 × 7 (ablation) | SparsePillars-SS · Waymo Open Dataset, 20% training data | vehicle 3D AP: 51.57 -> 56.77 | Improved |
| Increase regional grouping size from 3.20m to 3.84m (ablation) | SST · Waymo Open Dataset, 20% training data | vehicle LEVEL_1 AP: 66.9 -> 67.9 | Improved |
| Increase regional grouping size from 3.20m to 5.12m (ablation) | SST · Waymo Open Dataset, 20% training data | pedestrian LEVEL_1 AP: 70.4 -> 71.1 | Improved |
| Increase SRA block count from 5 to 7 (ablation) | SST · Waymo Open Dataset, 20% training data | vehicle LEVEL_1 AP: 67.7 -> 67.8 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- PV-RCNN++ — PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detection (CVPR 2020) · better on 0 of 2 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
- MsSVT: Mixed-scale Sparse Voxel Transformer for 3D Object Detection on Point Clouds (NeurIPS 2022 · similarity 0.90)
- Fully Sparse 3D Object Detection (NeurIPS 2022 · similarity 0.89)
- Super Sparse 3D Object Detection (TPAMI 2023 · similarity 0.89)
- PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 2024 · similarity 0.88)
- An LSTM Approach to Temporal 3D Object Detection in LiDAR Point Clouds (ECCV 2020 · similarity 0.88)
- SAFDNet: A Simple and Effective Network for Fully Sparse 3D Object Detection (CVPR 2024 · similarity 0.87)
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.
