Research map / Point clouds / Segmentation and detection
SEED: A Simple and Effective 3D DETR in Point Clouds
Zhe Liu, Jinghua Hou, Xiaoqing Ye, Tong Wang, Jingdong Wang, Xiang Bai
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 heatmap-based query selection with dual query selection (DQS) (ablation) | SEED-S · Waymo val (20% train) | mAPH L2: 64.6 -> 67.4 | Improved |
| replace box attention with deformable grid attention (DGA) (ablation) | SEED-S · Waymo val (20% train) | mAPH L2: 64.6 -> 66.4 | Improved |
| combine DQS and DGA (ablation) | SEED-S · Waymo val (20% train) | mAPH L2: 64.6 -> 68.2 | Improved |
| use DQS instead of learnable or Top-N query selection (ablation) | SEED-S · Waymo val (20% train) | mAPH L2: 66.6 -> 68.2 (learnable); 66.8 -> 68.2 (Top-N) | Improved |
| use heatmap-based query selection (ablation) | SEED-S · Waymo val (20% train) | mAPH L2: 68.2 -> 65.0 | Hurt |
| use deformable grid attention instead of global attention (ablation) | SEED-S · Waymo val (20% train) | feasibility: OOM -> 68.2 mAPH L2 | Improved |
| use deformable grid attention instead of deformable or box attention (ablation) | SEED-S · Waymo val (20% train) | mAPH L2: 67.5 -> 68.2 | Improved |
| use quality-aware Hungarian matching instead of traditional matching (ablation) | SEED-S · Waymo val (20% train) | mAPH L2: 67.6 -> 68.2 | Improved |
| use SEED-L over DSVT-Voxel on Waymo val | Waymo val single frame · Waymo val (100% train) | mAPH L2: 72.1 -> 73.5 | Improved |
| use SEED instead of TransFusion-L on nuScenes with the same 3D backbone | nuScenes val · nuScenes val | NDS / mAP: 70.1 -> 71.2 NDS; 65.1 -> 66.6 mAP | Improved |
| use higher-resolution SEED-L over SEED-B | Pedestrian detection · Waymo val (100% train) | Pedestrian APH L2: 70.8 -> 71.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Uni3DETR — Uni3DETR: Unified 3D Detection Transformer (NeurIPS 2023) · 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
- V-DETR: DETR with Vertex Relative Position Encoding for 3D Object Detection (ICLR 2024 · similarity 0.91)
- ConQueR: Query Contrast Voxel-DETR for 3D Object Detection (CVPR 2023 · similarity 0.90)
- Point2Seq: Detecting 3D Objects As Sequences (CVPR 2022 · similarity 0.89)
- AS-Det: Active Sampling for Adaptive 3D Object Detection in Point Clouds (AAAI 2025 · similarity 0.89)
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.88)
- Uni3DETR: Unified 3D Detection Transformer (NeurIPS 2023 · 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.
