List your company
ResearchFormerly WizWand

OOD Detection on CIFAR-10 (IND) vs LSUN C (OOD)

99.98AUROCMedix
40.980856.297971.61586.9321Oct 18, 2021Jun 16, 2022Feb 13, 2023Oct 13, 2023Jun 10, 2024Feb 7, 2025Oct 7, 2025
Updated 3mo ago

Evaluation results

MethodLinks
MedixTraining Data=Pin and Pwild2025.1099.98——0.01
ASHTraining Data=Pin only2025.1099.73——0.9
PALMMethod type=Training, Augmentations=Strong, Auxiliary outlier data=false, Backbone=ResNet-182024.0599.65——121
MixOEBackbone=ResNet-182024.0599.61——14
CSITraining Data=Pin only2025.1099.55——1.95
Energy + pNMLBackbone=WideResNet-402021.1099.598.697.1—
DivOEBackbone=ResNet-182024.0599.5——188
GradPCA-BatchBackbone=ResNetV2-50 (BiT-M)2025.0599.45——2.23
KNNBackbone=ResNetV2-50 (BiT-M)2025.0599.43——2.15
EnergyBackbone=WideResNet-402021.1099.498.697—
Hopfield BoostingBackbone=ResNet-182024.0599.4——82
Hopfield BoostingMethod type=OE, Augmentations=Weak, Auxiliary outlier data=true, Backbone=ResNet-182024.0599.4——82
KNN+Training Data=Pin only2025.1099.39——2.95
SSD+Method type=Training, Augmentations=Strong, Auxiliary outlier data=false, Backbone=ResNet-182024.0599.37——283
DOEBackbone=ResNet-182024.0599.3——322
Energy (w/ OE)Training Data=Pin and Pwild2025.1099.3——2.35
WOODSTraining Data=Pin and Pwild2025.1099.14——0.31
DALBackbone=ResNet-182024.0599.13——417
DOSBackbone=ResNet-182024.0599.04——366
NPOSMethod type=Training, Augmentations=Strong, Auxiliary outlier data=false, Backbone=ResNet-182024.0598.97——552
Kernel PCA (CoRP)Backbone=ResNetV2-50 (BiT-M)2025.0598.91——5.07
GradPCA (block 4)Backbone=ResNetV2-50 (BiT-M)2025.0598.9——6.01
POEMBackbone=ResNet-182024.0598.89——402
GAIA-ZBackbone=ResNet-34 (TIMM)2025.0598.87——5.27
MSP-OEBackbone=ResNet-182024.0598.83——702
GradPCA-VecBackbone=ResNetV2-50 (BiT-M)2025.0598.76——6.59
DICETraining Data=Pin only2025.1098.68——6.32
MahalanobisBackbone=ResNetV2-50 (BiT-M)2025.0598.65——6.06
EBO-OEBackbone=ResNet-182024.0598.43——682
EnergyTraining Data=Pin only2025.1098.35——8.26
OETraining Data=Pin and Pwild2025.1098.16——1.91
GAIA-ABackbone=ResNet-34 (TIMM)2025.0597.53——13.45
GradPCA (block 3)Backbone=ResNet-34 (TIMM)2025.0597.49——11.06
ASHMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0597.33——1,313
NCIBackbone=ResNetV2-50 (BiT-M)2025.0597.32——10.26
GradPCABackbone=ResNetV2-50 (BiT-M)2025.0597.21——15.1
ReActTraining Data=Pin only2025.1097.21——14.38
ODINTraining Data=Pin only2025.1097.04——15.52
GradPCA+DICEBackbone=ResNetV2-50 (BiT-M)2025.0596.78——17.95
EBOMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0596.73——1,725
MaxLogitMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0596.52——1,850
GENMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0596.48——1,940
GradPCA-BatchBackbone=ResNet-34 (TIMM)2025.0595.87——25.22
GradPCA-VecBackbone=ResNet-34 (TIMM)2025.0595.75——26.09
NCIBackbone=ResNet-34 (TIMM)2025.0595.69——20.51
MSPTraining Data=Pin only2025.1095.65——30.8
KNNTraining Data=Pin only2025.1095.26——25.55
Proj. GradsBackbone=ResNetV2-50 (BiT-M)2025.0595.09——21.48
Revisited PCABackbone=ResNet-34 (TIMM)2025.0594.95——20.8
KNNBackbone=ResNet-34 (TIMM)2025.0594.93——31.81
GAIA-ABackbone=ResNetV2-50 (BiT-M)2025.0594.67——24.36
EnergyBackbone=ResNet-34 (TIMM)2025.0594.46——21.57
Max logitsBackbone=ResNet-34 (TIMM)2025.0594.36——22.45
ODINBackbone=ResNet-34 (TIMM)2025.0594.36——22.45
Kernel PCA (CoRP)Backbone=ResNet-34 (TIMM)2025.0594.27——36.72
MahalanobisTraining Data=Pin only2025.1094.15——39.22
DICEBackbone=ResNet-34 (TIMM)2025.0593.83——20.52
GradPCABackbone=ResNet-34 (TIMM)2025.0593.81——31.29
Revisited PCABackbone=ResNetV2-50 (BiT-M)2025.0593.75——24.36
MSPBackbone=ResNet-34 (TIMM)2025.0593.11——39.86
ReActBackbone=ResNet-34 (TIMM)2025.0592.89——21.71
GradOrthBackbone=ResNet-34 (TIMM)2025.0592.57——34.75
MahalanobisBackbone=ResNet-34 (TIMM)2025.0590.4——59.66
Proj. GradsBackbone=ResNet-34 (TIMM)2025.0590.04——30.16
GradPCA+DICEBackbone=ResNet-34 (TIMM)2025.0590.03——42.09
MSPBackbone=ResNetV2-50 (BiT-M)2025.0589.45——29.7
MSPMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0588.37——4,582
ODINBackbone=ResNetV2-50 (BiT-M)2025.0584.09——37.92
Max logitsBackbone=ResNetV2-50 (BiT-M)2025.0584.08——37.91
GAIA-ZBackbone=ResNetV2-50 (BiT-M)2025.0582.62——51.87
EnergyBackbone=ResNetV2-50 (BiT-M)2025.0581.06——46.99
GradOrthBackbone=ResNetV2-50 (BiT-M)2025.0580.69——81.39
ReActBackbone=ResNetV2-50 (BiT-M)2025.0546.36——98.03
DICEBackbone=ResNetV2-50 (BiT-M)2025.0543.25——99.74