Image Classification on CIFAR10 (test) (Robustness & Clean Accuracy)
92.97Robust AccuracyCore-tuning
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Core-tuningAttack type=l2, Epsilon=0.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 92.97 | 96.82 | |
| Core-tuningAttack type=l2, Epsilon=1.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 92.32 | 96.9 | |
| Core-tuningAttack type=l2, Epsilon=2.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 92.05 | 96.87 | |
| AT-CE-ConAttack type=l2, Epsilon=0.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 90.74 | 94.71 | |
| AT-CE-ConAttack type=l2, Epsilon=1.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 90.29 | 94.8 | |
| AT-CE-ConAttack type=l2, Epsilon=2.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 89.7 | 94.27 | |
| AT-CEAttack type=l2, Epsilon=1.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 89.6 | 94.28 | |
| AT-CEAttack type=l2, Epsilon=2.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 89.16 | 94.15 | |
| Core-tuningAttack type=linf, Epsilon=2/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 86.92 | 96.29 | |
| AT-CEAttack type=l2, Epsilon=0.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 86.59 | 92 | |
| AT-CE-ConAttack type=linf, Epsilon=2/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 85.07 | 94.56 | |
| AT-CEAttack type=linf, Epsilon=2/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 83.2 | 93.05 | |
| Core-tuningAttack type=linf, Epsilon=4/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 82.01 | 95.95 | |
| AT-CE-ConAttack type=linf, Epsilon=4/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 79.75 | 93.79 | |
| AT-CEAttack type=linf, Epsilon=4/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 75.82 | 91.99 | |
| Core-tuningAttack type=linf, Epsilon=8/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 74.83 | 95.9 | |
| AT-CE-ConAttack type=linf, Epsilon=8/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 70.7 | 93.38 | |
| AT-CEAttack type=linf, Epsilon=8/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 69.27 | 92.79 | |
| CEAttack type=l2, Epsilon=0.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 50.25 | 94.7 | |
| CEAttack type=l2, Epsilon=1.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 48.29 | 94.7 | |
| CEAttack type=l2, Epsilon=2.5, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 46.82 | 94.7 | |
| CEAttack type=linf, Epsilon=2/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 25.13 | 94.7 | |
| CEAttack type=linf, Epsilon=4/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 12.28 | 94.7 | |
| CEAttack type=linf, Epsilon=8/255, Attack method=PGD-10, Backbone=ResNet-50, Pre-training=MoCo-v22021.02 | 4.57 | 94.7 |