Out-of-Distribution Detection on ImageNet-1K OpenOOD Far-OOD v1.5
15.62FPR@95ISH (Ours)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ISH (Ours)Backbone=ResNet50, Training Strategy=ISH, Epochs=90+10, Postprocessor=SCALE2023.09 | 15.62 | 96.79 | 76.74 | |
| ResNet50 TorchVision ModelBackbone=ResNet50, Training Strategy=TorchVision Model, Epochs=90, Postprocessor=SCALE2023.09 | 16.53 | 96.53 | 76.13 | |
| ResNet50 TorchVision Model ExtendedBackbone=ResNet50, Training Strategy=TorchVision Model Extended, Epochs=90+10, Postprocessor=SCALE2023.09 | 18.48 | 96.24 | 76.84 | |
| RegMixupBackbone=ResNet50, Training Strategy=RegMixup, Epochs=90+30, Postprocessor=SCALE2023.09 | 19.87 | 95.94 | 76.88 | |
| AugMixBackbone=ResNet50, Training Strategy=AugMix, Epochs=180, Postprocessor=SCALE2023.09 | 21.01 | 95.99 | 77.64 | |
| CIDERTraining Strategy=CIDER, Epochs=90+30, Postprocessor=KNN2023.09 | 28.69 | 92.18 | — | |
| LogitNormTraining Strategy=LogitNorm, Epochs=90+30, Postprocessor=MSP2023.09 | 31.33 | 91.54 | 76.45 |