Enhance the Visual Representation via Discrete Adversarial Training

Open paper
Benchmark Model Rank Results
domain-generalization-on-imagenet-aMAE+DAT (ViT-H)#10Top-1 accuracy %: 68.92
domain-generalization-on-imagenet-cMAE+DAT (ViT-H)#3mean Corruption Error (mCE): 31.4Number of params: 632M
domain-generalization-on-imagenet-rMAE+DAT (ViT-H)#11Top-1 Error Rate: 34.39
domain-generalization-on-imagenet-sketchMAE+DAT (ViT-H)#10Top-1 accuracy: 50.03
domain-generalization-on-stylized-imagenetMAE+DAT (ViT-H)#1Top 1 Accuracy: 32.77
image-classification-on-imagenetMAE+DAT (ViT-H)#108Top 1 Accuracy: 87.02%

Rank counts only results with a code link.