Research map / Architectures and efficiency / Attention and transformers
DeiT III: Revenge of the ViT
Hugo Touvron, Matthieu Cord, Hervé Jégou
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 |
|---|---|---|---|
| use binary cross-entropy instead of cross-entropy (ablation) | ViT-B · ImageNet-1k val | top-1 accuracy: 80.9 -> 82.2 | Improved |
| use binary cross-entropy instead of cross-entropy (ablation) | ViT-S · ImageNet-1k val | top-1 accuracy: 79.8 -> 79.8 | No clear effect |
| add LayerScale (ablation) | ViT-B · ImageNet-1k val | top-1 accuracy: 82.2 -> 82.5 | Improved |
| add LayerScale (ablation) | ViT-S · ImageNet-1k val | top-1 accuracy: 79.8 -> 80.1 | Improved |
| replace RandAugment with 3-Augment (ablation) | ViT-B · ImageNet-1k val | top-1 accuracy: 82.7 -> 83.1 | Improved |
| replace RandAugment with 3-Augment (ablation) | ViT-L · ImageNet-1k val | top-1 accuracy: 84.0 -> 84.2 | Improved |
| replace Auto-Augment with 3-Augment on ResNet-50 (ablation) | ResNet-50 · ImageNet-1k val | top-1 accuracy: 79.7 -> 79.4 | Hurt |
| add grayscale only to color jitter (ablation) | ViT-B · ImageNet-1k val | top-1 accuracy: 81.4 -> 81.0 | Hurt |
| add solarization to jitter, grayscale and blur (ablation) | ViT-B · ImageNet-1k val | top-1 accuracy: 82.7 -> 83.1 | Improved |
| add color jitter 0.3 to 3-Augment (ablation) | ViT-B · ImageNet-1k val | top-1 accuracy: 83.1 -> 83.1 | No clear effect |
| use Simple Random Crop instead of Random Resized Crop (ablation) | ViT-S · ImageNet-1k val (21k pretrain) | top-1 accuracy: 81.6 -> 82.1 | Improved |
| use Simple Random Crop instead of Random Resized Crop (ablation) | ViT-S · ImageNet-1k val (1k only) | top-1 accuracy: -0.9% | Hurt |
Closest papers by meaning
- An Empirical Study of Training Self-Supervised Vision Transformers (ICCV 2021 · similarity 0.93)
- A Closer Look at Self-Supervised Lightweight Vision Transformers (ICML 2023 · similarity 0.91)
- Tokens-to-Token ViT: Training Vision Transformers From Scratch on ImageNet (ICCV 2021 · similarity 0.91)
- DeiT-LT: Distillation Strikes Back for Vision Transformer Training on Long-Tailed Datasets (CVPR 2024 · similarity 0.90)
- Vision Transformers Need Registers (ICLR 2024 · similarity 0.90)
- Vision Transformers with Self-Distilled Registers (NeurIPS 2025 · similarity 0.89)
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.
