Research map / Detection and segmentation / CNN features
DETRs With Hybrid Matching
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 |
|---|---|---|---|
| Add a parallel one-to-many matching branch during training while retaining one-to-one matching for inference | Deformable-DETR · COCO val, 12 epochs, R50 | AP: 47.0 -> 48.7 | Helped |
| Add a parallel one-to-many matching branch during training | PETRv2 · nuScenes val, 36 epochs | NDS: 50.68 -> 52.38 | Helped |
| Add a parallel one-to-many matching branch during training | PETR · COCO val, 100 epochs, R50 | AP: 69.3 -> 70.9 | Helped |
| Add a parallel one-to-many matching branch during training | TransTrack · MOT17 val | MOTA: 67.1 -> 68.7 | Helped |
| Use one-to-many matching for the whole training process instead of the hybrid branch scheme (ablation) | Deformable-DETR · COCO 2017 val, 12 epochs | AP: 48.7 -> 49.4 | Helped |
| Use one-to-many matching for the whole training process (ablation) | Deformable-DETR · COCO 2017 val, 12 epochs | inference FPS: 6.7 -> 5.3 | Hurt |
| Reduce the repeated-target factor from K=6 to K=1 (ablation) | H-Deformable-DETR · COCO 2017 val, 12 epochs | AP: 48.6 -> 46.4 | Hurt |
| Increase one-to-many query count from T=300 to T=1500 (ablation) | H-Deformable-DETR · COCO 2017 val, 12 epochs | AP: 47.8 -> 48.7 | Helped |
| Use an independent encoder for the one-to-many branch rather than sharing the encoder (ablation) | H-Deformable-DETR · COCO 2017 val, 12 epochs | AP: 48.7 -> 47.3 | Hurt |
| Use independent decoder layers for the one-to-many branch (ablation) | H-Deformable-DETR · COCO 2017 val, 12 epochs | AP: 48.7 -> 48.3 | Hurt |
| Use independent box and classification heads across branches (ablation) | H-Deformable-DETR · COCO 2017 val, 12 epochs | AP: 48.7 -> 48.5 | No clear effect |
| Set one-to-many loss weight lambda to 0.1 instead of 1 (ablation) | H-Deformable-DETR · COCO 2017 val, 12 epochs | AP: 48.7 -> 47.8 | Hurt |
Closest papers by meaning
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.90)
- CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection (AAAI 2025 · similarity 0.87)
- DEIM: DETR with Improved Matching for Fast Convergence (CVPR 2025 · similarity 0.87)
- Order Matters: Competition-Guided Query Ordering for RNN-Based Object Detection (NeurIPS 2026 · similarity 0.87)
- Hybrid Proposal Refiner: Revisiting DETR Series from the Faster R-CNN Perspective (CVPR 2024 · similarity 0.87)
- DETRs Beat YOLOs on Real-time 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.