Image Classification on CIFAR-10 (Robustness and Corruption Analysis)
92.28Accuracy (Clean)VGG16
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| VGG16Training Protocol=Standard Training2025.11 | 92.28 | 74.53 | 89.08 | 93.36 | 88.76 | 0.49 | 0.31 | 0 | 84.9 | 76.56 | 99.9 | 93.84 | |
| WRN50Training Protocol=Standard Training2025.11 | 91.13 | 81.32 | 94.2 | 96.25 | 94.02 | 6.94 | 7.24 | 0.01 | 88.98 | 77.3 | 100.26 | 96.13 | |
| ResNet50Training Protocol=Standard Training2025.11 | 90.75 | 86.84 | 92.42 | 94.94 | 92.23 | 5.8 | 6.05 | 0 | 88.02 | 76.45 | 100.11 | 95.93 | |
| ResNet18Training Protocol=Standard Training2025.11 | 87.72 | 76.29 | 95.28 | 97.21 | 95.13 | 3.1 | 3.3 | 0 | 83.63 | 73.65 | 99.66 | 97.31 | |
| WRN50Training Protocol=Adversarial Training (PGD-10)2025.11 | 78.83 | 96.33 | 99.86 | 99.97 | 99.87 | 58.33 | 55.8 | 50.35 | 92.83 | 88.16 | 98.2 | 98.95 | |
| ResNet50Training Protocol=Adversarial Training (PGD-10)2025.11 | 75.29 | 95.66 | 99.66 | 99.8 | 99.65 | 59.74 | 56.74 | 51.27 | 92.92 | 89.27 | 97.84 | 98.75 | |
| ResNet18Training Protocol=Adversarial Training (PGD-10)2025.11 | 72.42 | 95.59 | 99.78 | 99.84 | 99.76 | 60.67 | 56.55 | 51.53 | 95.02 | 89.34 | 98.34 | 99.35 |