Research map / Point clouds / Segmentation and detection
SEFormer: Structure Embedding Transformer for 3D Object Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 vanilla Transformer with SEFormer in both the SSM and detection head (ablation) | Waymo vehicle detection · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 76.10 -> 79.02 | Helped |
| add SEFormer to the SSM while retaining a Transformer head (ablation) | Waymo vehicle detection · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 76.10 -> 77.54 | Helped |
| replace the Transformer head with SEFormer after using SEFormer in the SSM (ablation) | Waymo vehicle detection · Waymo Open val, vehicle LEVEL_1 | 3D mAP: +1.48% | Helped |
| use 2 parallel SEFormer blocks rather than 1 or 3 (ablation) | SEFormer spatial structure module · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 78.76 -> 79.02; 78.81 -> 79.02 | Helped |
| use 2 attention heads rather than 1 or 4 (ablation) | SEFormer · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 78.87 -> 79.02; 79.00 -> 79.02 | Helped |
| add conv3 features to conv1 (ablation) | SEFormer multi-scale feature inputs · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 78.54 -> 78.81 | Helped |
| add conv2 to conv1 + conv3 + conv4 (ablation) | SEFormer multi-scale feature inputs · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 79.02 -> 78.77 | Hurt |
| use grid interpolation instead of random radius sampling (ablation) | SEFormer key-point sampling · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 78.63 -> 79.02 | Helped |
| use half-parallel, half-chained SSM rather than fully parallel or fully chained SSM (ablation) | SEFormer spatial structure module · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 78.70 -> 79.02; 78.86 -> 79.02 | Helped |
| use SEFormer rather than the strongest reported baseline | Waymo vehicle 3D detection · Waymo Open val, vehicle LEVEL_1 | 3D mAP: 77.82 -> 79.02 | Helped |
| use SEFormer rather than Voxel-to-Point | Waymo vehicle 3D detection · Waymo Open val, vehicle LEVEL_1, 0-30m | 3D mAP: 93.23 -> 93.10 | Hurt |
| use SEFormer rather than Voxel-to-Point | Waymo vehicle 3D detection · Waymo Open val, vehicle LEVEL_1, 30-50m | 3D mAP: 76.21 -> 78.07 | Helped |
Closest papers by meaning
- 3D Object Detection With Pointformer (CVPR 2021 · similarity 0.90)
- DOPS: Learning to Detect 3D Objects and Predict Their 3D Shapes (CVPR 2020 · similarity 0.89)
- Seg2Box: 3D Object Detection by Point-Wise Semantics Supervision (AAAI 2025 · similarity 0.87)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 2024 · similarity 0.87)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.86)
- PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 2024 · 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.