Image Classification on Tiny-ImageNet augmented with 1x10^6 EDM-generated images (train)
65.23CleantrTRADES + LORE
Evaluation results
| Method | Links | |
|---|---|---|
| TRADES + LOREBackbone=PreResNet18 (Swish), Adversarial Training Method=TRADES, Boosting Algorithm=LORE-boosted, Training Data Augmentation=1M EDM images2024.03 | 65.23 | |
| TRADESBackbone=PreResNet18 (Swish), Adversarial Training Method=TRADES, Boosting Algorithm=Vanilla, Training Data Augmentation=1M EDM images2024.03 | 64.91 | |
| TRADES + SLOREBackbone=PreResNet18 (Swish), Adversarial Training Method=TRADES, Boosting Algorithm=SLORE-boosted, Training Data Augmentation=1M EDM images2024.03 | 64.76 | |
| MART + SLOREBackbone=PreResNet18 (Swish), Adversarial Training Method=MART, Boosting Algorithm=SLORE-boosted, Training Data Augmentation=1M EDM images2024.03 | 55.17 | |
| MART + LOREBackbone=PreResNet18 (Swish), Adversarial Training Method=MART, Boosting Algorithm=LORE-boosted, Training Data Augmentation=1M EDM images2024.03 | 53.04 | |
| MARTBackbone=PreResNet18 (Swish), Adversarial Training Method=MART, Boosting Algorithm=Vanilla, Training Data Augmentation=1M EDM images2024.03 | 52.52 |