Adversarial Robustness Evaluation on CIFAR-100 (test)
74.86Clean Accuracy AvgNatural
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| NaturalBackbone=MobileNet-v22023.12 | 74.86 | 53.9 | 0.15 | 5.95 | 0 | 1.423 | 0 | 0 | — | 0 | 0 | — | 0 | 0 | — | |
| BATBackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 65.39 | 38.7 | 0.228 | 33.31 | 5.3 | 0.577 | 27.08 | 2.9 | 0.683 | 22.8 | 1.5 | 0.79 | 21.28 | 1.1 | 0.823 | |
| CFABackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 59.15 | 34.1 | 0.255 | 32.5 | 8.8 | 0.551 | 29.38 | 7 | 0.601 | 25.16 | 4 | 0.69 | 23.78 | 3.2 | 0.727 | |
| Fair-ARDBackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 58.97 | 31.4 | 0.289 | 33.76 | 9.3 | 0.542 | 30.07 | 7.3 | 0.607 | 27.64 | 5.6 | 0.651 | 25.79 | 4 | 0.7 | |
| RSLADBackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 58.71 | 27.1 | 0.311 | 34.3 | 8.3 | 0.565 | 30.58 | 6.3 | 0.621 | 28.11 | 4.3 | 0.683 | 26.32 | 3.3 | 0.724 | |
| TRADESBackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 57.1 | 30 | 0.284 | 31.7 | 8.2 | 0.584 | 29.43 | 6.2 | 0.618 | 25.25 | 3.7 | 0.716 | 24.39 | 3.2 | 0.739 | |
| SATBackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 56.7 | 22.9 | 0.338 | 32.1 | 7 | 0.58 | 28.61 | 5.3 | 0.65 | 26.55 | 3.5 | 0.706 | 24.36 | 2.4 | 0.764 | |
| ABSLDBackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 56.66 | 32 | 0.252 | 33.87 | 12.8 | 0.448 | 31.24 | 11.4 | 0.489 | 26.41 | 7.5 | 0.564 | 24.57 | 6.7 | 0.597 | |
| FRLBackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 56.3 | 26.5 | 0.311 | 31.03 | 8 | 0.548 | 27.52 | 6.1 | 0.602 | 25.4 | 3.7 | 0.657 | 23.28 | 2.7 | 0.714 | |
| AdaADBackbone=MobileNet-v2, Perturbation Budget (epsilon)=8/2552023.12 | 54.59 | 24.5 | 0.343 | 31.4 | 6.1 | 0.616 | 27.96 | 4.8 | 0.677 | 25.72 | 2.1 | 0.752 | 23.8 | 1.6 | 0.807 |