Patch Classification on OpenSRH (test)
83.1Clean Accuracy[17]
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| [17]Training=from Scratch, Model=ResNet-502025.03 | 83.1 | 82.29 | 10.25 | 9.84 | 10.27 | 9.85 | 11.15 | 10.34 | 5.52 | 5.01 | 5.57 | 5.02 | 5.97 | 5.31 | |
| [17]Training=from ImageNet weights, Model=ResNet-502025.03 | 80.07 | 77.78 | 23.61 | 22.11 | 23.32 | 21.84 | 21.76 | 21.03 | 13.07 | 12.8 | 13.06 | 12.75 | 12.53 | 12.71 | |
| [17]Training=from ImageNet weights, Model=WideResNet-502025.03 | 79.24 | 76.46 | 32.28 | 26.53 | 32.24 | 26.51 | 28.95 | 23.86 | 20.87 | 17.21 | 20.9 | 17.25 | 19.24 | 15.97 | |
| HSATTraining=from ImageNet weights, Model=WideResNet-502025.03 | 68.35 | 67.76 | 60.51 | 59.27 | 60.54 | 59.32 | 61.44 | 60.19 | 53.13 | 51.26 | 53.12 | 51.26 | 54.44 | 52.73 | |
| HSATTraining=from ImageNet weights, Model=ResNet-502025.03 | 66.48 | 66.33 | 64.07 | 61.36 | 64.11 | 61.4 | 65.11 | 62.4 | 56.97 | 55 | 56.98 | 55.04 | 57.92 | 56.06 | |
| HSATTraining=from Scratch, Model=ResNet-502025.03 | 66.43 | 65.61 | 64.07 | 61.36 | 64.11 | 61.4 | 65.11 | 62.4 | 57.18 | 53.48 | 57.1 | 53.38 | 58.52 | 54.76 | |
| HSAT-PatchTraining=from ImageNet weights, Model=ResNet-502025.03 | 50.8 | 44.36 | 51.08 | 44.95 | 51.06 | 44.94 | 51.1 | 44.96 | 50.29 | 45.22 | 50.29 | 45.22 | 50.41 | 45.31 | |
| HSAT-PatchTraining=from Scratch, Model=ResNet-502025.03 | 44.84 | 39.63 | 44.67 | 39.65 | 44.67 | 39.65 | 44.8 | 39.8 | 44.23 | 39.76 | 44.22 | 39.75 | 44.26 | 39.88 |