Image Classification on CIFAR-100 (Corruption and Perturbation Robustness)
72.02Accuracy (Clean)WRN50
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| WRN50Training Protocol=Standard Training2025.11 | 72.02 | 58.88 | 82.88 | 88.24 | 82.58 | 4.56 | 4.94 | 0.01 | 77.2 | 59.5 | 99.89 | 89.03 | |
| VGG16Training Protocol=Standard Training2025.11 | 71.02 | 52.13 | 75.86 | 83.48 | 75.44 | 0.98 | 0.81 | 0 | 78.26 | 56.62 | 99.94 | 85.24 | |
| ResNet50Training Protocol=Standard Training2025.11 | 70.57 | 58.93 | 80.73 | 86.78 | 80.31 | 3.47 | 3.69 | 0.01 | 76.39 | 59.33 | 100.18 | 88.51 | |
| ResNet18Training Protocol=Standard Training2025.11 | 63.38 | 58.65 | 86.51 | 91.48 | 86.17 | 2.45 | 3.02 | 0.02 | 69.03 | 60.18 | 99.4 | 92.21 | |
| ResNet50Training Protocol=Adversarial Training (PGD-10)2025.11 | 52.14 | 85.88 | 100.06 | 100.09 | 100.07 | 40.26 | 38.01 | 32.83 | 88.7 | 81.4 | 98.85 | 97.1 | |
| ResNet18Training Protocol=Adversarial Training (PGD-10)2025.11 | 49.39 | 89.54 | 99.96 | 99.95 | 99.92 | 43.88 | 39.89 | 34.64 | 90.61 | 81.8 | 97.67 | 97.89 | |
| WRN50Training Protocol=Adversarial Training (PGD-10)2025.11 | 21.1 | 94.28 | 100.11 | 100 | 100.03 | 57.44 | 45.73 | 40.28 | 100 | 90.43 | 96.35 | 99.62 |