Adversarial Robustness Accuracy on Mini-ImageNet (test)
76.37Clean AccuracyNatural Training
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Natural TrainingEnvironment=Gray-box2026.06 | 76.37 | 13.4 | 3.39 | 2.03 | 1.19 | 1.07 | |
| MAPEEnvironment=Gray-box2026.06 | 74.86 | 63.62 | 54.42 | 55.89 | 53.53 | 51.29 | |
| Mixup InferenceEnvironment=Gray-box2026.06 | 73.81 | 42.73 | 37.7 | 35.48 | 32.45 | 29.98 | |
| LDTEnvironment=Gray-box2026.06 | 73.76 | 55.32 | 46.77 | 46.87 | 44.8 | 41.58 | |
| HGDEnvironment=Gray-box2026.06 | 72.41 | 51.72 | 40.81 | 38.77 | 35.78 | 33.38 | |
| Pixel DeflectionEnvironment=Gray-box2026.06 | 67.28 | 41.04 | 35.88 | 34.71 | 34.37 | 33.67 | |
| TVMEnvironment=Gray-box2026.06 | 66.17 | 50.85 | 45.62 | 46.78 | 40.49 | 39.48 | |
| Feature DenoisingEnvironment=Gray-box2026.06 | 59 | 21.45 | 12.17 | 11.52 | 11.39 | 10.57 | |
| Adversarial TrainingEnvironment=Gray-box2026.06 | 58.39 | 32.27 | 29.22 | 29.42 | 29.87 | 29.04 |