Black-box Adversarial Attack on ImageNet
62.6Top-1 Accuracy (JPEG)NCF
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| NCFSubstitute Model=Inc-v32022.10 | 62.6 | — | 37.9 | 26 | 23.5 | 73.5 | 66.2 | 61.2 | 58.2 | 60.1 | 32.6 | 26.9 | |
| SAESubstitute Model=Inc-v32022.10 | 52.7 | — | 33.6 | 13 | 11.5 | 62.2 | 53.7 | 55.6 | 53.5 | 50.1 | 19.8 | 13.7 | |
| ColorFoolSubstitute Model=Inc-v32022.10 | 42.1 | — | 31.1 | 8.9 | 7.5 | 65.8 | 43.6 | 57.5 | 52.1 | 46.3 | 16.9 | 10.6 | |
| ACESubstitute Model=Inc-v32022.10 | 31.2 | — | 30.8 | 6.1 | 3.8 | 47.4 | 32.6 | 36.3 | 31.6 | 29.3 | 13.4 | 7.6 | |
| cAdvSubstitute Model=Inc-v32022.10 | 30.6 | — | 27.8 | 12.4 | 10.5 | 40.9 | 36.2 | 42.8 | 37.2 | 33.1 | 20.3 | 15.1 | |
| ReColorAdvSubstitute Model=Inc-v32022.10 | 26.9 | — | 34.4 | 7.8 | 5.6 | 38 | 27.6 | 32.9 | 28.8 | 24.3 | 14.7 | 9.3 | |
| Clean baselineSubstitute Model=Inc-v32022.10 | 17.4 | — | 26.7 | 2.4 | 2 | 28.2 | 16.8 | 29.2 | 24.4 | 19.8 | 7.7 | 4.2 |