Research map / Architectures and efficiency / Attention and transformers
Dual Vision Transformer
Ting Yao, Yehao Li, Yingwei Pan, Yu Wang, Xiao-Ping Steven Zhang, Tao Mei
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 self-attention layer among semantic tokens in the semantic pathway (ablation) | Dual block (Dual-ViT-S) · ImageNet-1k val | top-1 accuracy | Improved |
| add feed-forward layer in the semantic pathway (ablation) | Dual block (Dual-ViT-S) · ImageNet-1k val | top-1 accuracy | Improved |
| apply self-attention before cross-attention in semantic pathway (ablation) | Dual block (Dual-ViT-S) · ImageNet-1k val | top-1 accuracy | Improved |
| add Token Labeling objective via MixToken | Dual-ViT-S · ImageNet-1k val (254x254) | top-1 accuracy: 83.4 -> 84.1 | Improved |
| use Dual-ViT-L with Token Labeling instead of VOLO-D4 | Dual-ViT-L · ImageNet-1k val (254x254) | top-1 accuracy: 85.7 | Improved |
| fine-tune at 384x384 resolution instead of 254x254 | Dual-ViT-S · ImageNet-1k val | top-1 accuracy: 83.4 -> 85.2 | Improved |
| replace PVTv2-B2 backbone with Dual-ViT-S | RetinaNet 1x · COCO val2017 | AP: 44.6 -> 46.2 | Improved |
| replace PVTv2-B2 backbone with Dual-ViT-S | Cascade Mask R-CNN · COCO val2017 | box AP: 51.1 -> 52.4 | Improved |
| replace Twins-SVT-S backbone with Dual-ViT-S | UPerNet · ADE20K val | mIoU: 47.1 -> 49.2 | Improved |
| replace Swin-S backbone with Dual-ViT-B | UPerNet · ADE20K val | mIoU: 49.5 -> 51.7 | Improved |
Closest papers by meaning
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.89)
- DaViT: Dual Attention Vision Transformers (ECCV 2022 · similarity 0.89)
- Unified Visual Transformer Compression (ICLR 2022 · similarity 0.88)
- A Survey on Efficient Vision Transformers: Algorithms, Techniques, and Performance Benchmarking (TPAMI 2024 · similarity 0.88)
- Multi-criteria Token Fusion with One-step-ahead Attention for Efficient Vision Transformers (CVPR 2024 · similarity 0.87)
- Skip-Attention: Improving Vision Transformers by Paying Less Attention (ICLR 2024 · 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.
