Research map / Architectures and efficiency / Attention and transformers
Vision Transformer With Quadrangle Attention
Qiming Zhang, Jing Zhang, Yufei Xu, Dacheng Tao
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 quadrangle attention in place of fixed windows (ablation) | ViT-B/16 window attention · ImageNet-1k val | top-1 accuracy: 81.2 -> 82.9 | Improved |
| add shear transformation on top of scale and shift (ablation) | QFormer_p-B · ImageNet-1k val | top-1 accuracy: 82.3 -> 82.5 | Improved |
| add rotation on top of scale, shift, shear (ablation) | QFormer_p-B · ImageNet-1k val | top-1 accuracy: 82.5 -> 82.6 | Improved |
| add projection on top of scale, shift, shear, rotation (ablation) | QFormer_p-B · ImageNet-1k val | top-1 accuracy: 82.6 -> 82.9 | Improved |
| use quadrangle attention instead of window attention with 2 full attention layers (ablation) | ViT-B/16 · ImageNet-1k val | top-1 accuracy: 82.5 -> 83.0 | Improved |
| use quadrangle attention instead of window attention with 6 full attention layers (ablation) | ViT-B/16 · ImageNet-1k val | top-1 accuracy: 82.9 -> 83.3 | Improved |
| compose basic transformations instead of predicting the matrix directly (ablation) | QA in ViT-B/16 · ImageNet-1k val | top-1 accuracy: 82.7 -> 82.9 | Improved |
| quadrangle attention instead of shifted window attention | ViT-B/16 · ImageNet-1k val | top-1 accuracy: 82.0 -> 82.9 | Improved |
| add the regularization term (lambda 0 to 0.1) (ablation) | QFormer_p-B · ImageNet-1k val | top-1 accuracy: diverged -> 82.9 | Improved |
| increase regularization weight lambda from 1 to 10 (ablation) | QFormer_p-B · ImageNet-1k val | top-1 accuracy: 82.9 -> 82.7 | Hurt |
| replace window attention with quadrangle attention | ViT-B + UPerNet · ADE20k val | mIoU: 39.7 -> 43.6 | Improved |
| replace Swin-T windows with quadrangle attention | Swin-T Mask RCNN 1x · COCO val | box mAP: 43.7 -> 45.9 | Improved |
Closest papers by meaning
- QT-ViT: Improving Linear Attention in ViT with Quadratic Taylor Expansion (NeurIPS 2024 · similarity 0.85)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.85)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.84)
- LookHere: Vision Transformers with Directed Attention Generalize and Extrapolate (NeurIPS 2024 · similarity 0.84)
- MimiQ: Low-Bit Data-Free Quantization of Vision Transformers with Encouraging Inter-Head Attention Similarity (AAAI 2025 · similarity 0.84)
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers (NeurIPS 2025 · 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.
