Image Classification on ImageNet-C extra corruptions (test)
49.6Average AccuracyPRIME
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PRIMEBackbone=ResNet-502021.12 | 49.6 | 59.5 | 61.4 | 60.1 | 48.1 | 26.9 | 28.3 | 66.5 | 36.4 | 41.9 | 66.5 | |
| DABackbone=ResNet-502021.12 | 48.3 | 60.1 | 61.1 | 57.7 | 46.8 | 25.4 | 24.4 | 68.4 | 26.5 | 45.6 | 66.8 | |
| AugMixBackbone=ResNet-502021.12 | 46.5 | 59.5 | 56.5 | 59.1 | 51.7 | 25.6 | 21.6 | 65.3 | 23.1 | 36.2 | 66.4 | |
| StandardBackbone=ResNet-502021.12 | 40 | 36.2 | 57.8 | 54.1 | 46.1 | 14.4 | 20.9 | 61.6 | 24.3 | 19 | 65.2 |