Research map / Detection and segmentation / CNN features
Deformable DETR: Deformable Transformers for End-to-End Object Detection
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 |
|---|---|---|---|
| Replace DETR attention with deformable attention | DETR · COCO 2017 val | AP: 42.0 -> 43.8 | Helped |
| Replace DETR attention with deformable attention | DETR · COCO 2017 val | AP_S: 20.5 -> 26.4 | Helped |
| Add iterative bounding box refinement (ablation) | Deformable DETR · COCO 2017 val | AP: 43.8 -> 45.4 | Helped |
| Add two-stage region proposals (ablation) | Deformable DETR with iterative refinement · COCO 2017 val | AP: 45.4 -> 46.2 | Helped |
| Add FPN to multi-scale deformable attention (ablation) | Deformable DETR · COCO 2017 val | AP: 43.8 -> 43.8 | No clear effect |
| Replace FPN with BiFPN (ablation) | Deformable DETR · COCO 2017 val | AP: 43.8 -> 43.9 | No clear effect |
| Add multi-scale inputs (ablation) | Deformable DETR with single-scale attention and K=1 · COCO 2017 val | AP: 39.7 -> 41.4 | Helped |
| Increase sampling points from K=1 to K=4 (ablation) | Deformable DETR with multi-scale inputs, no multi-scale attention · COCO 2017 val | AP: 41.4 -> 42.3 | Helped |
| Add multi-scale attention (ablation) | Deformable DETR with multi-scale inputs and K=4 · COCO 2017 val | AP: 42.3 -> 43.8 | Helped |
| Replace multi-scale deformable attention with single-scale deformable convolution-like attention (ablation) | Deformable DETR · COCO 2017 val | AP: 43.8 -> 39.7 | Hurt |
| Use ResNet-101 instead of ResNet-50 | Deformable DETR · COCO 2017 test-dev | AP: 46.9 -> 48.7 | Helped |
| Use ResNeXt-101 instead of ResNet-101 | Deformable DETR · COCO 2017 test-dev | AP: 48.7 -> 49.0 | Helped |
Closest papers by meaning
- Sparse DETR: Efficient End-to-End Object Detection with Learnable Sparsity (ICLR 2022 · similarity 0.92)
- Hybrid Proposal Refiner: Revisiting DETR Series from the Faster R-CNN Perspective (CVPR 2024 · similarity 0.88)
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.87)
- DETRs Beat YOLOs on Real-time Object Detection (CVPR 2024 · similarity 0.86)
- Not All Tokens Matter All The Time: Dynamic Token Aggregation Towards Efficient Detection Transformers (ICML 2025 · similarity 0.86)
- RDD: Robust Feature Detector and Descriptor using Deformable Transformer (CVPR 2025 · 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.