Research map / Architectures and efficiency / Attention and transformers
Vision Transformer With Deformable Attention
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 |
|---|---|---|---|
| Replace the last two stages' attention with deformable attention | DAT-T · ImageNet-1K | Top-1 accuracy: 81.3 -> 82.0 (+0.7) | Helped |
| Replace the last two stages' attention with deformable attention | DAT-S · ImageNet-1K | Top-1 accuracy: 83.0 -> 83.7 (+0.7) | Helped |
| Replace the last two stages' attention with deformable attention | DAT-B · ImageNet-1K | Top-1 accuracy: 83.5 -> 84.0 (+0.5) | Helped |
| Replace the last two stages' attention with deformable attention | DAT-B · ImageNet-1K val, 384² | Top-1 accuracy: 84.5 -> 84.8 (+0.3) | Helped |
| Use deformable attention in place of the Swin-T backbone attention | RetinaNet · COCO val, 1x | AP: 41.7 -> 42.8 | Helped |
| Use deformable attention in place of the Swin-T backbone attention | RetinaNet · COCO val, 3x | AP: 44.8 -> 45.6 | Helped |
| Use deformable attention in place of the Swin-T backbone attention | Mask R-CNN · COCO val, 3x | box AP: 46.0 -> 47.1 | Helped |
| Use deformable attention in place of the Swin-B backbone attention | Cascade Mask R-CNN · COCO val, 3x | box AP: 51.9 -> 53.0 | Helped |
| Use deformable attention in place of the Swin-T backbone attention | UperNet · ADE20K val | mIoU: 44.51 -> 45.54 | Helped |
| Remove spatial offsets from deformable attention (ablation) | DAT-T · ImageNet-1K | Top-1 accuracy: 82.0 -> 81.7 | Hurt |
| Add deformable relative position embedding (ablation) | DAT-T · ImageNet-1K | Top-1 accuracy: 81.7 -> 82.0 | Helped |
| Replace deformable relative position embedding with fixed position bias (ablation) | DAT-T · ImageNet-1K | Top-1 accuracy: 82.0 -> 81.8 | Hurt |
Closest papers by meaning
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.89)
- Breaking the Low-Rank Dilemma of Linear Attention (CVPR 2025 · similarity 0.88)
- Skip-Attention: Improving Vision Transformers by Paying Less Attention (ICLR 2024 · similarity 0.88)
- LookHere: Vision Transformers with Directed Attention Generalize and Extrapolate (NeurIPS 2024 · similarity 0.87)
- Bridging the Divide: Reconsidering Softmax and Linear Attention (NeurIPS 2024 · similarity 0.87)
- Attention-Only Transformers via Unrolled Subspace Denoising (ICML 2025 · 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.