Image Classification on ImageNet-100-C Alternative Corruptions
65.9Average AccuracyPRIME
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PRIMEBackbone=ResNet-18, Training Epochs=1.5x2021.12 | 65.9 | 77.1 | 75.6 | 74.1 | 59.4 | 54 | 46.3 | 77.6 | 50.4 | 67.7 | 76.4 | |
| PRIMEBackbone=ResNet-182021.12 | 65 | 74.9 | 74.3 | 73.2 | 59.2 | 53.4 | 47.5 | 76.8 | 48.6 | 66.9 | 75.5 | |
| DABackbone=ResNet-182021.12 | 63.8 | 77.1 | 76.6 | 72.6 | 60.9 | 42.9 | 44.3 | 78 | 43.4 | 64.5 | 77.8 | |
| AugMixBackbone=ResNet-182021.12 | 61 | 63 | 73.2 | 69.4 | 39.9 | 44.9 | 77.4 | 42.8 | 44.7 | 79.8 | — | |
| StandardBackbone=ResNet-182021.12 | 55.1 | 47.6 | 71.3 | 70.1 | 66.4 | 29.5 | 45.7 | 72.1 | 34.6 | 34.9 | 78.4 |