Image Classification on ImageNet-1K (Metrics: Nat, Union)
66.1Nat AccuracyE-AT + RoME
Evaluation results
| Method | Links | ||
|---|---|---|---|
| E-AT + RoMEArchitecture=XCiT-S, Pre-training=l_inf-adversarial training2026.07 | 66.1 | 27.6 | |
| MAX + RoMEArchitecture=XCiT-S, Pre-training=l_inf-adversarial training2026.07 | 65.1 | 30.5 | |
| E-ATArchitecture=XCiT-S, Pre-training=l_inf-adversarial training2026.07 | 65 | 25.9 | |
| RAMP + RoMEArchitecture=XCiT-S, Pre-training=l_inf-adversarial training2026.07 | 64.2 | 32.1 | |
| MAXArchitecture=XCiT-S, Pre-training=l_inf-adversarial training2026.07 | 64.1 | 27.2 | |
| RAMPArchitecture=XCiT-S, Pre-training=l_inf-adversarial training2026.07 | 62.8 | 29.4 |