Image Classification on Caltech101 (PGD-l2 attack, ε=0.5, test)
89.21Robust AccuracyCore-tuning
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Core-tuningBackbone=ResNet-50, Pre-training=MoCo-v2, Attack Type=PGD-10, Perturbation Norm=l2, Epsilon=0.52021.02 | 89.21 | 92.83 | |
| AT-CE-ConBackbone=ResNet-50, Pre-training=MoCo-v2, Attack Type=PGD-10, Perturbation Norm=l2, Epsilon=0.52021.02 | 88.67 | 92.61 | |
| AT-CEBackbone=ResNet-50, Pre-training=MoCo-v2, Attack Type=PGD-10, Perturbation Norm=l2, Epsilon=0.52021.02 | 87.35 | 91.61 | |
| CEBackbone=ResNet-50, Pre-training=MoCo-v2, Attack Type=PGD-10, Perturbation Norm=l2, Epsilon=0.52021.02 | 55.69 | 91.87 |