Research map / Point clouds / Segmentation and detection
Graph Neural Network and Spatiotemporal Transformer Attention for 3D Video Object Detection From Point Clouds
Junbo Yin, Jianbing Shen, Xin Gao, David Crandall, Ruigang Yang
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 |
|---|---|---|---|
| replace concatenation with GMPNet for short-term encoding (ablation) | CenterPoint-PointPillars · nuScenes val (CenterPoint-PointPillars) | mAP: 48.25 -> 49.78 | Improved |
| replace ConvGRU with AST-GRU for long-term aggregation (ablation) | ConvGRU · nuScenes val (CenterPoint-PointPillars) | mAP: 56.54 -> 57.96 | Improved |
| add spatial transformer attention (STA) to ConvGRU (ablation) | ConvGRU · nuScenes val (CenterPoint-PointPillars) | mAP: 56.54 -> 57.54 | Improved |
| add temporal transformer attention (TTA) on top of STA (ablation) | STA-GRU · nuScenes val (CenterPoint-PointPillars) | mAP: 57.54 -> 57.96 | Improved |
| use tracklet fusion instead of concatenation (ablation) | concatenation · nuScenes val (CenterPoint-PointPillars) | mAP: 49.35 -> 50.09 | Improved |
| use 3D ConvNet instead of concatenation (ablation) | concatenation · nuScenes val (CenterPoint-PointPillars) | mAP: 49.35 -> 52.84 | Improved |
| add GMPNet to AST-GRU (full online model) (ablation) | AST-GRU · nuScenes val (CenterPoint-PointPillars) | mAP: 57.96 -> 58.78 | Improved |
| use offline mode with future keyframes (ablation) | Full Model (online) · nuScenes val (CenterPoint-PointPillars) | mAP: 58.78 -> 59.37 | Improved |
| add motion map input to TTA (ablation) | TTA · nuScenes val (TTA analysis) | mAP: 26.03 -> 27.28 | Improved |
| use two TTA layers rather than one (ablation) | TTA · nuScenes val (TTA analysis) | mAP: 26.55 -> 27.28 | Improved |
| use three TTA layers rather than two (ablation) | TTA · nuScenes val (TTA analysis) | mAP: 27.28 -> 26.42 | Hurt |
| use video (VID) model instead of single-frame PointPillars | PointPillars · nuScenes test | NDS: 45.3 -> 53.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- 3DSSD — 3DSSD: Point-Based 3D Single Stage Object Detector (CVPR 2020) · better on 2 of 2 shared results
- PointPillars — PointPillars: Fast Encoders for Object Detection From Point Clouds (CVPR 2019) · better on 2 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
- LiDAR-Based Online 3D Video Object Detection With Graph-Based Message Passing and Spatiotemporal Transformer Attention (CVPR 2020 · similarity 0.94)
- Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud (CVPR 2020 · similarity 0.91)
- Graph R-CNN: Towards Accurate 3D Object Detection with Semantic-Decorated Local Graph (ECCV 2022 · similarity 0.90)
- AGO-Net: Association-Guided 3D Point Cloud Object Detection Network (TPAMI 2021 · similarity 0.90)
- PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 2024 · similarity 0.89)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 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.
