Research map / Detection and segmentation / CNN features
Group DETR: Fast DETR Training with Group-Wise One-to-Many Assignment
Qiang Chen, Xiaokang Chen, Jian Wang, Shan Zhang, Kun Yao, Haocheng Feng, Junyu Han, Errui Ding, Gang Zeng, Jingdong Wang
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 |
|---|---|---|---|
| add group-wise one-to-many assignment with K groups of queries (ablation) | Conditional DETR · COCO val2017 (12 epochs) | mAP: 32.6 -> 34.8 | Improved |
| add separate self-attention per group (ablation) | Group DETR · COCO val2017 (12 epochs) | mAP: 34.8 -> 37.6 | Improved |
| use naive one-to-many assignment instead of group-wise (ablation) | Conditional DETR · COCO val2017 (12 epochs) | mAP: 32.6 -> 8.4 | Hurt |
| increase number of groups K up to 11 (ablation) | Group DETR · COCO val2017 (12 epochs) | mAP | Improved |
| apply Group DETR to the encoder and decoder rather than decoder only (ablation) | Conditional DETR · COCO val2017 (50 epochs) | mAP: 41.5 -> 42.9 | Improved |
| give baseline longer training to match Group DETR time (ablation) | Conditional DETR · COCO val2017 | mAP: 32.6 -> 34.4 vs Group DETR 37.6 | No clear effect |
| add Group DETR to Conditional DETR (12 epochs) | Conditional DETR · COCO val2017 (12 epochs) | mAP: 32.6 -> 37.6 | Improved |
| add Group DETR to DINO-4scale | DINO · COCO val2017 | mAP: 49.4 -> 50.1 (12e); 50.9 -> 51.3 (36e) | Improved |
| add Group DETR to DINO-4scale-Swin-L | DINO Swin-L · COCO val2017 (36 epochs) | mAP: 58.0 -> 58.4 | Improved |
| gains shrink with longer training and stronger baselines | Group DETR · COCO val2017 | mAP gain: Conditional DETR +5.0 (12e) -> +2.5 (50e); DN-DETR-DC5 +2.6 -> +0.5 | Mixed |
| combine Group DETR with DN-DETR denoising queries | DAB-DETR · COCO val2017 (12 epochs) | mAP: +1.5 mAP over Group DETR | Improved |
| increase DN-DETR denoising queries (ablation) | DN-DETR · COCO val2017 (12 epochs) | mAP | No clear effect |
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.
- PETR v2 — PETRv2: A Unified Framework for 3D Perception from Multi-Camera Images (ICCV 2023) · better on 1 of 1 shared result
- PETR — PETR: Position Embedding Transformation for Multi-View 3D Object Detection (ECCV 2022) · 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
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.91)
- Mr. DETR: Instructive Multi-Route Training for Detection Transformers (CVPR 2025 · similarity 0.90)
- Mr. DETR++: Instructive Multi-Route Training for Detection Transformers with Mixture-of-Experts (TPAMI 2026 · similarity 0.90)
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.88)
- Salience DETR: Enhancing Detection Transformer with Hierarchical Salience Filtering Refinement (CVPR 2024 · similarity 0.87)
- Hybrid Proposal Refiner: Revisiting DETR Series from the Faster R-CNN Perspective (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.
