Research map / Detection and segmentation / CNN features
Quadtree Attention for Vision Transformers
Shitao Tang, Jiahui Zhang, Siyu Zhu, Ping Tan
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 |
|---|---|---|---|
| Replace LoFTR linear attention with QuadTree-B | LoFTR-lite · ScanNet | AUC@20°: 49.0 -> 51.6 | Improved |
| Replace LoFTR linear attention with QuadTree-B using K=16 and ViT-like blocks | LoFTR · ScanNet | AUC@20°: 56.6 -> 61.6 | Improved |
| Replace STTR standard transformer with QuadTree-B | STTR · Scene Flow FlyingThings3D | EPE (px): 0.45 px vs 0.46 px | No clear effect |
| Replace PVTv2-b2 spatial-reduction attention with QuadTree-B | PVTv2-b2 · ImageNet-1K val | Top-1 accuracy (%): 82.0 -> 82.7 | Improved |
| Replace PVTv2-b2 spatial-reduction attention with QuadTree-B | PVTv2-b2 · COCO val2017 (RetinaNet) | AP: 44.6 -> 46.2 | Improved |
| Use QuadTree-B rather than QuadTree-A (ablation) | QuadTree attention · COCO val2017 (RetinaNet) | AP: 37.0 -> 38.4 | Improved |
| Add multiscale position encoding (ablation) | QuadTree-B-b2 · ImageNet-1K | Top-1 accuracy (%): 82.6 -> 82.7 | Improved |
| Add multiscale position encoding (ablation) | QuadTree-B-b2 · COCO val2017 (RetinaNet) | AP: 44.9 -> 46.2 | Improved |
| Increase K from 1 to 32 (ablation) | QuadTree-B · COCO val2017 (RetinaNet) | AP: 37.3 -> 38.5 | Improved |
| Increase K from 1 to 16 (ablation) | QuadTree-B · ScanNet (LoFTR-lite) | AUC@20°: 48.9 -> 51.7 | Improved |
| Use QuadTree-B instead of QuadTree-A (ablation) | QuadTree attention · ScanNet feature matching | AUC@20°: 50.5 -> 51.6 | Improved |
Closest papers by meaning
- Vision Transformer With Quadrangle Attention (TPAMI 2024 · similarity 0.90)
- Vicinity Vision Transformer (TPAMI 2023 · similarity 0.87)
- CAT: Circular-Convolutional Attention for Sub-Quadratic Transformers (NeurIPS 2025 · similarity 0.87)
- QuadMamba: Learning Quadtree-based Selective Scan for Visual State Space Model (NeurIPS 2024 · similarity 0.86)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.85)
- PADRe: A Unifying Polynomial Attention Drop-in Replacement for Efficient Vision Transformer (ICLR 2025 · similarity 0.85)
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.
