Image Classification on ImageNet (val) (Adversarial Robustness - FGSM/PGD)
70.67FGSM Top-1 AccuracyViT-B/16†
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ViT-B/16†Attention=QUEST, Epochs=400 + 20, Training Strategy=DeiT-32026.03 | 70.67 | 1.194 | 54.64 | 2.033 | — | — | |
| ViT-B/16†Attention=Standard, Epochs=400 + 20, Training Strategy=DeiT-32026.03 | 69.41 | 1.229 | 52.55 | 2.08 | — | — | |
| ViT-S/16Attention=QUEST, Epochs=2002026.03 | 67.02 | 1.328 | 57.51 | 1.886 | 40.98 | 2.54 | |
| ViT-S/16Attention=Standard, Epochs=2002026.03 | 65.9 | 1.362 | 54.47 | 2.05 | 38.22 | 2.699 | |
| ViT-L/16Attention=QUEST, Epochs=1002026.03 | 62.85 | 1.669 | 53.54 | 2.316 | — | — | |
| ViT-L/16Attention=QKNorm-LN, Epochs=1002026.03 | 58.95 | 1.849 | 51.79 | 2.355 | — | — |