Research map / Detection and segmentation / 3D detection and driving
Spatial-Temporal Graph Enhanced DETR Towards Multi-Frame 3D Object Detection
Yifan Zhang, Zhiyu Zhu, Junhui Hou, Dapeng Oliver Wu
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 |
|---|---|---|---|
| process multi-frame input as sequence with ConvGRU instead of concatenation (ablation) | STEMD · Waymo val L2 (4 frames) | mAPH: 69.62 -> 70.10 | Improved |
| add spatial-temporal graph attention network (ablation) | STEMD · Waymo val L2 (4 frames) | mAPH: 70.10 -> 72.15 | Improved |
| add temporal query recollection (ablation) | STEMD · Waymo val L2 (4 frames) | mAPH: 72.15 -> 72.87 | Improved |
| add IoU regularization term (ablation) | STEMD · Waymo val L2 (4 frames) | mAPH: 72.87 -> 73.75 | Improved |
| use STGA-Net instead of conventional multi-head self-attention (ablation) | STEMD · Waymo val L2 (4 frames) | mAPH: 71.85 -> 73.75 | Improved |
| use feature-based learned edge weights instead of distance-based (ablation) | STGA-Net · Waymo val L2 | mAPH: 71.17 -> 73.75 | Improved |
| add both spatial self-attention and temporal cross-attention (ablation) | STGA-Net · Waymo val L2 | mAPH: 71.56 -> 73.75 | Improved |
| add only temporal cross-attention (ablation) | STGA-Net · Waymo val L2 | mAPH: 71.56 -> 71.69 | Improved |
| remove edge limit (no limit versus 5 edges per node) (ablation) | STGA-Net · Waymo val L2 | mAPH: 71.74 -> 73.75 | Improved |
| use 16-frame STEMD versus concatenation baseline (ablation) | STEMD · Waymo val L2 (16 frames) | mAPH: 69.65 -> 74.56 | Improved |
| increase concatenated frames from 8 to 16 (ablation) | Concat baseline · Waymo val L2 | mAPH: 69.68 -> 69.65 | No clear effect |
| recollect 300 queries from last frame (ablation) | STEMD · Waymo val L2 | mAPH: 73.12 -> 73.75 | Improved |
Closest papers by meaning
- Graph R-CNN: Towards Accurate 3D Object Detection with Semantic-Decorated Local Graph (ECCV 2022 · similarity 0.89)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.88)
- State Space Model Meets Transformer: A New Paradigm for 3D Object Detection (ICLR 2025 · similarity 0.88)
- MI-DETR: An Object Detection Model with Multi-time Inquiries Mechanism (CVPR 2025 · similarity 0.88)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.87)
- V-DETR: DETR with Vertex Relative Position Encoding for 3D Object Detection (ICLR 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.
