Research map / Architectures and efficiency / Attention and transformers
TokenMix: Rethinking Image Mixing for Data Augmentation in Vision Transformers
Jihao Liu, Boxiao Liu, Hang Zhou, Hongsheng Li, Yu Liu
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 CutMix with TokenMix | DeiT-S · ImageNet-1K val | top-1 accuracy (%): 79.8 -> 80.8 | Improved |
| replace CutMix with TokenMix | Swin-T · ImageNet-1K val | top-1 accuracy (%): 81.2 -> 81.6 | Improved |
| add content-based target refinement to CutMix (ablation) | DeiT-S · ImageNet-1K val | top-1 accuracy (%): 79.8 -> 80.5 | Improved |
| add target refinement to CutMix (ablation) | ResNet50 · ImageNet-1K val | top-1 accuracy (%): 79.3 -> 79.8 | Improved |
| use region-based mask instead of token blocks (ablation) | DeiT-S · ImageNet-1K val | top-1 accuracy (%): 80.6 -> 79.8 | Hurt |
| use block-based mask instead of random token mask on ResNet50 (ablation) | ResNet50 · ImageNet-1K val | top-1 accuracy (%): 78.3 -> 79.7 | Improved |
| add refinement to block-based masks (ablation) | DeiT-T · ImageNet-1K val | top-1 accuracy (%): 72.7 -> 73.2 | Improved |
| add refinement to random token masks (ablation) | DeiT-S · ImageNet-1K val | top-1 accuracy (%): 80.6 -> 80.6 | No clear effect |
| use weaker teacher for activation maps (NFNet-F6 to ResNet26) (ablation) | TokenMix DeiT-S · ImageNet-1K val | top-1 accuracy (%): 80.8 -> 80.5 | Hurt |
| use hand-crafted saliency instead of a neural teacher (ablation) | TokenMix DeiT-S · ImageNet-1K val | top-1 accuracy (%): 80.8 -> 80.1 | Hurt |
| add BCE loss instead of CE (ablation) | TokenMix DeiT-S · ImageNet-1K val | top-1 accuracy (%): 80.3 -> 80.8 | Improved |
| add BCE loss instead of CE (ablation) | CutMix DeiT-S · ImageNet-1K val | top-1 accuracy (%): 79.8 -> 79.8 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports TokenMix as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- SMMix: Self-Motivated Image Mixing for Vision Transformers (ICCV 2023 · better on 1 of 2 shared results)
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
- SMMix: Self-Motivated Image Mixing for Vision Transformers (ICCV 2023 · similarity 0.91)
- TokenMixup: Efficient Attention-guided Token-level Data Augmentation for Transformers (NeurIPS 2022 · similarity 0.91)
- MixPro: Data Augmentation with MaskMix and Progressive Attention Labeling for Vision Transformer (ICLR 2023 · similarity 0.90)
- Provable Benefit of Cutout and CutMix for Feature Learning (NeurIPS 2024 · similarity 0.87)
- Learning to Merge Tokens via Decoupled Embedding for Efficient Vision Transformers (NeurIPS 2024 · similarity 0.86)
- DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models (CVPR 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.
