Out-of-Distribution Detection on MNIST (ID) vs Fashion-MNIST (OOD) (test)
0.9955AUCDAPPr
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DAPPrBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9955 | — | |
| BNN-ARHTBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.9951 | 0.9947 | |
| BNN-ARHTIn-Distrib.=MNIST, Architecture=LeNet [22]2023.10 | 0.9951 | 0.9947 | |
| DPNBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.9941 | 0.9937 | |
| DPNIn-Distrib.=MNIST, Architecture=LeNet [22]2023.10 | 0.9941 | 0.9937 | |
| MC DropoutBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.9933 | 0.9927 | |
| MC DropoutIn-Distrib.=MNIST, Architecture=LeNet [22]2023.10 | 0.9933 | 0.9927 | |
| F-EDLBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9931 | — | |
| R-EDLBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9929 | — | |
| I-EDLBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9886 | — | |
| REGrad*layer-selective=true, perturbation=true2024.04 | 0.9878 | 0.9879 | |
| COMBOODBackbone=LeNet2026.02 | 0.9871 | 0.9863 | |
| PostNetBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.9859 | 0.947 | |
| PostNetIn-Distrib.=MNIST, Architecture=LeNet [22]2023.10 | 0.9859 | 0.947 | |
| I-EDLBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.9849 | 0.9889 | |
| KL-PNBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9816 | — | |
| Exgrad2024.04 | 0.9811 | 0.9798 | |
| EDLBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9808 | — | |
| Entropy2024.04 | 0.9804 | 0.9794 | |
| ExGrad V Term2024.04 | 0.9804 | 0.9795 | |
| Perturb θ2024.04 | 0.9799 | 0.9764 | |
| LA2024.04 | 0.9795 | 0.9737 | |
| DUQBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9792 | — | |
| MDSBackbone=LeNet2026.02 | 0.9761 | 0.9754 | |
| Reg. MahalanobisBackbone=LeNet2026.02 | 0.9757 | 0.9726 | |
| PostNetBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9724 | — | |
| Perturb x2024.04 | 0.9709 | 0.9539 | |
| Inserted Dropout2024.04 | 0.97 | 0.9575 | |
| MC DropoutBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.9656 | — | |
| GramBackbone=LeNet2026.02 | 0.9641 | 0.9637 | |
| KNNBackbone=LeNet2026.02 | 0.9568 | 0.951 | |
| UNGrad2024.04 | 0.9541 | 0.9491 | |
| MC-AA2024.04 | 0.9528 | 0.9524 | |
| ODINBackbone=LeNet2026.02 | 0.9491 | 0.9446 | |
| Kendall and GalBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.9254 | 0.9277 | |
| Kendall and GalIn-Distrib.=MNIST, Architecture=LeNet [22]2023.10 | 0.9254 | 0.9277 | |
| Deep EnsemblesBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.907 | 0.9108 | |
| Deep EnsemblesIn-Distrib.=MNIST, Architecture=LeNet [22]2023.10 | 0.907 | 0.9108 | |
| DetectronBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.7557 | 0.8375 | |
| DetectronIn-Distrib.=MNIST, Architecture=LeNet [22]2023.10 | 0.7557 | 0.8375 | |
| EDLBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | 0.7343 | 0.8022 | |
| EDLIn-Distrib.=MNIST, Architecture=LeNet [22]2023.10 | 0.7343 | 0.8022 | |
| RKL-PNBackbone=ConvNet (3 conv + 3 dense)2026.05 | 0.7218 | — | |
| NEGrad2024.04 | 0.4099 | 0.4441 | |
| GradNorm2024.04 | 0.2271 | 0.4088 | |
| RKL-PNBackbone=LeNet, In-Distribution Dataset=MNIST2023.10 | — | 0.7845 |