Image Classification on CIFAR-10 (test) (Cl-Acc/Rob-Acc)
92.68Classification Accuracy (Cl-Acc)AdjAttack
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AdjAttackModel=WideResNet-70-16, T=0.075, lp Attack=l_2, epsilon=0.52023.10 | 92.68 | 80.6 | — | |
| DiffAttackModel=WideResNet-70-16, T=0.075, lp Attack=l_2, epsilon=0.52023.10 | 92.68 | 70.31 | -10.29 | |
| AdjAttackModel=WideResNet-28-10, T=0.075, lp Attack=l_2, epsilon=0.52023.10 | 91.03 | 78.58 | — | |
| DiffAttackModel=WideResNet-28-10, T=0.075, lp Attack=l_2, epsilon=0.52023.10 | 91.03 | 64.06 | -14.52 | |
| AdjAttackModel=WideResNet-70-16, T=0.1, lp Attack=l_inf, epsilon=8/2552023.10 | 90.07 | 71.29 | — | |
| DiffAttackModel=WideResNet-70-16, T=0.1, lp Attack=l_inf, epsilon=8/2552023.10 | 90.07 | 45.31 | -25.98 | |
| AdjAttackModel=WideResNet-70-16, T=0.1, lp Attack=l_inf, epsilon=4/2552023.10 | 90.07 | 81.25 | — | |
| DiffAttackModel=WideResNet-70-16, T=0.1, lp Attack=l_inf, epsilon=4/2552023.10 | 90.07 | 75 | -6.25 | |
| AdjAttackModel=WideResNet-28-10, T=0.1, lp Attack=l_inf, epsilon=8/2552023.10 | 89.02 | 70.64 | — | |
| DiffAttackModel=WideResNet-28-10, T=0.1, lp Attack=l_inf, epsilon=8/2552023.10 | 89.02 | 46.88 | -23.76 | |
| AdjAttackModel=WideResNet-28-10, T=0.1, lp Attack=l_inf, epsilon=4/2552023.10 | 89.02 | 82.81 | — | |
| DiffAttackModel=WideResNet-28-10, T=0.1, lp Attack=l_inf, epsilon=4/2552023.10 | 89.02 | 71.88 | -10.93 |