Research map / Detection and segmentation / CNN features
DN-DETR: Accelerate DETR Training by Introducing Query DeNoising
Check 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 box and label denoising to DAB-DETR | DAB-DETR-R50 · COCO val2017, R50, 50 epochs | AP: 42.2 -> 44.1; +1.9 AP | Helped |
| Add denoising training to DAB-DETR | DAB-DETR-DC5-R50 · COCO val2017, 12 epochs | AP: 38.0 -> 41.7; +3.7 AP | Helped |
| Add denoising training to Deformable DETR | Deformable DETR-R50 · COCO val2017, multi-scale, 12 epochs | AP: 37.2 -> 43.4 | Helped |
| Add denoising training to other detection and segmentation models | Anchor DETR, Group-DAB-DETR, Faster R-CNN, SAM-DETR++, DINO, Vanilla DETR, Mask DINO, and Mask2Former · COCO val2017, 12 epochs where reported | AP: +1.2-2.6 AP; +0.5 AP; +1.6 AP; +1.4 AP; +2.0 AP; +0.7 AP; +1.0 AP | Helped |
| Remove label denoising (ablation) | DN-DETR · COCO val2017, R50, 50 epochs | AP: 43.4 -> 43.0 | Hurt |
| Add box denoising (ablation) | DN-DETR · COCO val2017, R50, 50 epochs | AP: 42.2 -> 43.4 | Helped |
| Remove the attention mask (ablation) | DN-DETR · COCO val2017, R50, 50 epochs | AP: 43.4 -> 24.0 | Hurt |
| Use 5 rather than 1 denoising group (ablation) | DN-DETR-R50 · COCO val2017, R50 | AP: 43.4 -> 44.1 | Helped |
| Use 5 rather than 1 denoising group (ablation) | DN-DETR-DC5-R50 · COCO val2017, DC5-R50 | AP: 45.6 -> 46.3 | Helped |
| Use denoising training for 25 epochs instead of baseline training for 50 epochs | DAB-DETR-DC5-R50 · COCO val2017, DC5-R50 | AP: 44.5 -> 44.4 | No clear effect |
| Increase noise scale too much (ablation) | DN-DETR · COCO val2017, R50 noise-scale experiment | AP | Hurt |
| Add center-shifting and box-scaling noise at suitable scales (ablation) | DN-DETR · COCO val2017, R50 noise-scale experiment | AP | Helped |
Closest papers by meaning
- AQ-DETR: Low-Bit Quantized Detection Transformer with Auxiliary Queries (AAAI 2024 · similarity 0.86)
- Order Matters: Competition-Guided Query Ordering for RNN-Based Object Detection (NeurIPS 2026 · similarity 0.85)
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.85)
- Salience DETR: Enhancing Detection Transformer with Hierarchical Salience Filtering Refinement (CVPR 2024 · similarity 0.84)
- DEIM: DETR with Improved Matching for Fast Convergence (CVPR 2025 · similarity 0.84)
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.84)
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.