Out-of-Distribution Detection on CIFAR-10 vs SVHN (Standard Test)
99.9AUROCSSD+
Evaluation results
| Method | Links | |
|---|---|---|
| SSD+2021.03 | 99.9 | |
| SSDk+k=52021.03 | 99.6 | |
| lambda (Similarity-only)2026.02 | 99.54 | |
| Gram Matrix2021.03 | 99.5 | |
| Contrastive + SupervisedBackbone=4x wider ResNet-50 network2021.03 | 99.5 | |
| SPEM-noiseFlow Model=ResFlow, Feature Extractor=ResNet-152, Hyperparameter alpha=0.12026.02 | 99.43 | |
| MahalanobisRequires additional out-of-distribution data for tuning=true2021.03 | 99.4 | |
| SPEMFlow Model=ResFlow, Feature Extractor=ResNet-152, Hyperparameter alpha=0.42026.02 | 99.13 | |
| Residual FlowsRequires additional out-of-distribution data for tuning=true2021.03 | 99.1 | |
| Rotation-loss + Supervised2021.03 | 98.9 | |
| Outlier exposure2021.03 | 98.4 | |
| CSI2021.03 | 97.9 | |
| CSI+LoRot-IForwarded Samples=280M2022.07 | 97.7 | |
| CSI+LoRot-EForwarded Samples=280M2022.07 | 97.5 | |
| SupCLRForwarded Samples=70M2022.07 | 97.3 | |
| CSIForwarded Samples=280M2022.07 | 96.5 | |
| ODINRequires additional out-of-distribution data for tuning=true2021.03 | 96.4 | |
| RotationsReproduced=true, Forwarded Samples=25M2022.07 | 96.1 | |
| Softmax-probs2021.03 | 95.9 | |
| LoRot-EForwarded Samples=5M2022.07 | 94.4 | |
| LIDFlow Model=ResFlow2026.02 | 93.6 | |
| LoRot-IForwarded Samples=5M2022.07 | 92.6 | |
| GMMFlow Model=ResFlow2026.02 | 89.16 | |
| SLA+SDReproduced=true, Forwarded Samples=20M2022.07 | 89.1 | |
| Typicality (Entropy)Flow Model=ResFlow2026.02 | 88.66 | |
| Likelihood RatioFlow Model=ResFlow2026.02 | 85.12 | |
| Cross EntropyReproduced=true, Forwarded Samples=5M2022.07 | 84.6 | |
| ComplexityFlow Model=ResFlow2026.02 | 79.43 | |
| CutmixReproduced=true, Forwarded Samples=5M2022.07 | 75.5 | |
| LikelihoodFlow Model=ResFlow2026.02 | 2.56 | |
| Typicality (sqrt(d))Flow Model=ResFlow2026.02 | 2.26 |