Research map / Architectures and efficiency / Attention and transformers
FLatten Transformer: Vision Transformer using Focused Linear Attention
Dongchen Han, Xuran Pan, Yizeng Han, Shiji Song, Gao Huang
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 focused function to vanilla linear attention (ablation) | FLatten-DeiT-T · ImageNet-1K val | Top-1 accuracy (%): 70.5 -> 71.8 | Improved |
| add depthwise convolution (DWC) rank restoration (ablation) | FLatten-DeiT-T · ImageNet-1K val | Top-1 accuracy (%): 71.8 -> 74.1 | Improved |
| vary focused factor p between 2 and 32 (ablation) | FLatten-Swin-T · ImageNet-1K val | Top-1 accuracy (%): 82.03, 82.11, 81.94, 81.99, 81.88 | No clear effect |
| enlarge attention window from 7^2 to 56^2 (ablation) | FLatten-Swin-T · ImageNet-1K val | Top-1 accuracy (%): 81.6 -> 82.1 | Improved |
| apply focused linear attention to first two stages instead of only stage 1 (ablation) | FLatten-Swin-T · ImageNet-1K val | Top-1 accuracy (%): 81.7 -> 82.1 | Improved |
| also apply focused linear attention in last two stages (ablation) | FLatten-Swin-T · ImageNet-1K val | Top-1 accuracy (%): 82.1 -> 81.9 | Hurt |
| use FLatten linear attention instead of other linear attention designs | DeiT-T setting · ImageNet-1K val | Top-1 accuracy (%): 72.9 -> 74.1 | Improved |
| replace Softmax attention with FLatten in DeiT-T | DeiT-T · ImageNet-1K val | Top-1 accuracy (%): 72.2 -> 74.1 | Improved |
| replace Swin attention with FLatten | Swin-T · ADE20K UperNet | mIoU: 44.51 -> 44.82 | Improved |
| replace attention with FLatten in larger models (CSwin-B at 384) | CSwin-B · ImageNet-1K val | Top-1 accuracy (%): 85.4 -> 85.5 | No clear effect |
Closest papers by meaning
- Breaking the Low-Rank Dilemma of Linear Attention (CVPR 2025 · similarity 0.90)
- Bridging the Divide: Reconsidering Softmax and Linear Attention (NeurIPS 2024 · similarity 0.90)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.89)
- PADRe: A Unifying Polynomial Attention Drop-in Replacement for Efficient Vision Transformer (ICLR 2025 · similarity 0.89)
- Skip-Attention: Improving Vision Transformers by Paying Less Attention (ICLR 2024 · similarity 0.88)
- MLP Can Be A Good Transformer Learner (CVPR 2024 · similarity 0.88)
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.
