Out-of-distribution image detection on Places365 (test)
28.2FPR95MSP
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MSPOOD Detector=Maximum Softmax Probability, Backbone=ResNet-18, Outlier Exposure=true2018.12 | 28.2 | 90.6 | 71 | |
| MOSBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=3.2, Fine-tuned=true2021.05 | 49.54 | 89.06 | — | |
| ENNTraining Dataset=CIFAR10, Scoring Function=softmax2021.07 | 51.09 | 87.56 | 96.76 | |
| Clipped HNNTraining Dataset=CIFAR10, Scoring Function=softmax2021.07 | 54.77 | 86.82 | 96.17 | |
| RankFeatBackbone=SqueezeNet, Block=Block 42022.09 | 61.31 | 80.52 | — | |
| RankFeatBackbone=SqueezeNet, Block=Block 3 + 42022.09 | 61.56 | 84.51 | — | |
| RankFeatBackbone=SqueezeNet, Block=Block 32022.09 | 62.94 | 85.82 | — | |
| MSPOOD Detector=Maximum Softmax Probability, Backbone=ResNet-18, Outlier Exposure=false2018.12 | 63.5 | 66.5 | 33.1 | |
| GradNormBackbone=SqueezeNet2022.09 | 65.99 | 83.28 | — | |
| RankFeatBackbone=T2T-ViT-242022.09 | 66.22 | 80.89 | — | |
| EnergyBackbone=SqueezeNet2022.09 | 67.74 | 82.73 | — | |
| ODINBackbone=T2T-ViT-242022.09 | 68.19 | 75.33 | — | |
| ReActBackbone=T2T-ViT-242022.09 | 68.93 | 78.2 | — | |
| MSPBackbone=T2T-ViT-242022.09 | 69.54 | 80.03 | — | |
| KL MatchingBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=20.6, Fine-tuned=true2021.05 | 72.61 | 76.49 | — | |
| EnergyBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=3.1, Fine-tuned=true2021.05 | 73.02 | 81.37 | — | |
| EnergyBackbone=T2T-ViT-242022.09 | 74.24 | 68.17 | — | |
| ODINBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=23.6, Fine-tuned=true2021.05 | 76.27 | 80.67 | — | |
| MSPBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=3.1, Fine-tuned=true2021.05 | 81.44 | 76.76 | — | |
| ODINBackbone=SqueezeNet2022.09 | 83.23 | 73.31 | — | |
| MSPBackbone=SqueezeNet2022.09 | 87.27 | 67 | — | |
| ReActBackbone=SqueezeNet2022.09 | 88.8 | 66.2 | — | |
| MahalanobisBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=145.4, Fine-tuned=true2021.05 | 89.75 | 64.46 | — | |
| MahalanobisBackbone=SqueezeNet2022.09 | 92.26 | 56.63 | — | |
| MahalanobisBackbone=T2T-ViT-242022.09 | 93.32 | 49.6 | — | |
| GradNormBackbone=T2T-ViT-242022.09 | 99.01 | 25.71 | — |