Research map / Architectures and efficiency / Attention and transformers
Doubly-Fused ViT: Fuse Information from Vision Transformer Doubly with Local Representation
Li Gao, Dong Nie, Bo Li, Xiaofeng Ren
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 Spatial Module (SM) to Context Module (ablation) | DFvT-S · ImageNet-1K val (ablation) | top-1 accuracy (%): 77.0 -> 77.5 | Improved |
| add DANE fusion to CM+SM (ablation) | DFvT-S · ImageNet-1K val (ablation) | top-1 accuracy (%): 77.5 -> 78.3 | Improved |
| add convolutional information flow to pre-SA fusion (ablation) | DFvT-S · ImageNet-1K val (ablation) | top-1 accuracy (%): 72.2 -> 78.3 | Improved |
| add max-pooling information flow to pre-SA fusion (ablation) | DFvT-S · ImageNet-1K val (ablation) | top-1 accuracy (%): 78.1 -> 78.3 | Improved |
| replace SUM fusion by DANE (ablation) | DFvT-S · ImageNet-1K val (ablation) | top-1 accuracy (%): 77.5 -> 78.3 | Improved |
| replace SE fusion by DANE (ablation) | DFvT-S · ImageNet-1K val (ablation) | top-1 accuracy (%): 78.0 -> 78.3 | Improved |
| reverse DANE (channel and spatial attention swapped between CM and SM) (ablation) | DFvT-S · ImageNet-1K val (ablation) | top-1 accuracy (%): 78.3 -> 78.1 | Hurt |
| downsample input feature map before SA (Downsampled ViT vs baseline ViT) (ablation) | window-SA ViT · ImageNet-1K val (design study) | top-1 accuracy (%): 74.7 -> 71.9 | Hurt |
| add parallel convolution path to Downsampled ViT (ablation) | Downsampled ViT · ImageNet-1K val (design study) | top-1 accuracy (%): 71.9 -> 76.9 | Improved |
| use DFvT-S design instead of Parallel ViT (ablation) | Parallel ViT · ImageNet-1K val (design study) | throughput (imgs/s): 1843.3 -> 2203.3 | Improved |
| downsample before SA instead of baseline ViT (ablation) | window-SA ViT · ImageNet-1K val (design study) | throughput (imgs/s): 1168.2 -> 3091.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MobileFormer-294M — Mobile-Former: Bridging MobileNet and Transformer (CVPR 2022) · better on 2 of 2 shared results
- ResT-Base — ResT: An Efficient Transformer for Visual Recognition (NeurIPS 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Survey on Efficient Vision Transformers: Algorithms, Techniques, and Performance Benchmarking (TPAMI 2024 · similarity 0.88)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.88)
- ViT-CoMer: Vision Transformer with Convolutional Multi-scale Feature Interaction for Dense Predictions (CVPR 2024 · similarity 0.88)
- LF-ViT: Reducing Spatial Redundancy in Vision Transformer for Efficient Image Recognition (AAAI 2024 · similarity 0.87)
- Multi-criteria Token Fusion with One-step-ahead Attention for Efficient Vision Transformers (CVPR 2024 · similarity 0.87)
- FasterViT: Fast Vision Transformers with Hierarchical Attention (ICLR 2024 · similarity 0.86)
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.
