Out-of-distribution detection on ImageNet-1k vs Texture
95.34AUROCViM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ViM2024.10 | 95.34 | 20.31 | — | |
| NNGuide2024.10 | 95.32 | 21.58 | — | |
| DINOv2+MLSBackbone=DINOv2, Training=Linear Probe, Postprocessor=MLS2024.10 | 91.82 | 33.95 | — | |
| ViT-B+CE+RMDSBackbone=ViT-B, Training=Cross Entropy (CE), Postprocessor=RMDS2024.10 | 89.38 | 37.22 | — | |
| NPOSIn-distribution image requirement=Required2024.09 | 88.8 | — | 46.12 | |
| HFTTIn-distribution image requirement=Not required2024.09 | 88.23 | — | 43.08 | |
| MCMIn-distribution image requirement=Not required2024.09 | 87.52 | — | 50.9 | |
| KNNIn-distribution image requirement=Required2024.09 | 85.67 | — | 64.35 | |
| GradNormMethod Space=Gradient, Backbone=ResNetv2-101, Pre-trained on=ImageNet-1k, Input Resolution=480x4802021.10 | 81.07 | 61.42 | — | |
| MSPIn-distribution image requirement=Required2024.09 | 79.33 | — | 67.01 | |
| ZOCIn-distribution image requirement=Not required2024.09 | 76.46 | — | 98.9 | |
| ODINMethod Space=Output, Backbone=ResNetv2-101, Pre-trained on=ImageNet-1k, Input Resolution=480x4802021.10 | 76.3 | 81.31 | — | |
| EnergyMethod Space=Output, Backbone=ResNetv2-101, Pre-trained on=ImageNet-1k, Input Resolution=480x4802021.10 | 75.79 | 80.87 | — | |
| MaxLogitIn-distribution image requirement=Not required2024.09 | 75.63 | — | 68.17 | |
| MSPMethod Space=Output, Backbone=ResNetv2-101, Pre-trained on=ImageNet-1k, Input Resolution=480x4802021.10 | 74.45 | 82.73 | — | |
| EnergyIn-distribution image requirement=Not required2024.09 | 72.36 | — | 51.24 | |
| MahalanobisMethod Space=Feature, Backbone=ResNetv2-101, Pre-trained on=ImageNet-1k, Input Resolution=480x4802021.10 | 72.1 | 52.23 | — | |
| MahalanobisIn-distribution image requirement=Required2024.09 | 64.23 | — | 98.46 |