Pyramid Adversarial Training Improves ViT

Reported on 6 benchmarks across 2 tasks · 1 paper

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Methodology3 results

Domain AdaptationonImageNet-R
Top-1 Error Rate· 2021-11-30
46.08
best: 3.9 (Model soups (BASIC-L))
Pyramid Adversarial Training Improves ViT Performance arXiv:2111.15121
Domain AdaptationonImageNet-C
mean Corruption Error (mCE)· 2021-11-30
41.42
best: 22 (EfficientNet-L2+RPL)
Pyramid Adversarial Training Improves ViT Performance arXiv:2111.15121
Domain AdaptationonImageNet-Sketch
Top-1 accuracy· 2021-11-30
41.04
best: 77.18 (Model soups (BASIC-L))
Pyramid Adversarial Training Improves ViT Performance arXiv:2111.15121

Computer Vision3 results

Domain GeneralizationonImageNet-R
Top-1 Error Rate· 2021-11-30
46.08
best: 3.9 (Model soups (BASIC-L))
Pyramid Adversarial Training Improves ViT Performance arXiv:2111.15121
Domain GeneralizationonImageNet-C
mean Corruption Error (mCE)· 2021-11-30
41.42
best: 28.2 (DINOv2 (ViT-g/14, frozen model, linear eval))
Pyramid Adversarial Training Improves ViT Performance arXiv:2111.15121
Domain GeneralizationonImageNet-Sketch
Top-1 accuracy· 2021-11-30
41.04
best: 77.18 (Model soups (BASIC-L))
Pyramid Adversarial Training Improves ViT Performance arXiv:2111.15121