List your company
ResearchFormerly WizWand

OOD Detection on CIFAR-10 (IND) vs iSUN (OOD)

100AUROCDOE
36.809653.214869.6286.0252Oct 18, 2021May 24, 2022Dec 29, 2022Aug 4, 2023Mar 10, 2024Oct 14, 2024May 21, 2025
Updated 8mo ago

Evaluation results

MethodLinks
DOEBackbone=ResNet-182024.05100——0
DOSBackbone=ResNet-182024.0599.99——0
EBO-OEBackbone=ResNet-182024.0599.98——0
Hopfield BoostingBackbone=ResNet-182024.0599.97——0
Hopfield BoostingMethod type=OE, Augmentations=Weak, Auxiliary outlier data=true, Backbone=ResNet-182024.0599.97——0
MSP-OEBackbone=ResNet-182024.0599.96——0
DALBackbone=ResNet-182024.0599.93——0
DivOEBackbone=ResNet-182024.0599.88——0
MixOEBackbone=ResNet-182024.0599.87——0.17
POEMBackbone=ResNet-182024.0599.87——0
Energy + pNMLBackbone=WideResNet-402021.1099.498.797—
EnergyBackbone=WideResNet-402021.1099.398.396.7—
ODINBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0898.9——3.98
ASH-SBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0898.9——5.17
GradPCA-BatchBackbone=ResNetV2-50 (BiT-M)2025.0598.74——5.55
VRA-PBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0898.69——5.7
ASH-BBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0898.45——8.59
GradOrthBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0898.32——4.25
ASH-PBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0898.29——8.46
Energy scoreBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0898.07——10.07
KNNBackbone=ResNetV2-50 (BiT-M)2025.0598.03——9.49
ReActBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0897.72——12.72
DICEBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0897.55——4.36
GAIA-ZBackbone=ResNet-34 (TIMM)2025.0597.47——12.56
MahalanobisBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0897.25——9.78
MahalanobisBackbone=ResNetV2-50 (BiT-M)2025.0597.13——16.24
NCIBackbone=ResNetV2-50 (BiT-M)2025.0596.95——15.94
GradPCA-VecBackbone=ResNetV2-50 (BiT-M)2025.0596.84——17.05
GAIA-ABackbone=ResNet-34 (TIMM)2025.0596.66——19.95
GradPCA (block 4)Backbone=ResNetV2-50 (BiT-M)2025.0596.49——19.87
NCIBackbone=ResNet-34 (TIMM)2025.0596.41——13.04
GradPCA (block 3)Backbone=ResNet-34 (TIMM)2025.0596.32——20.92
NPOSMethod type=Training, Augmentations=Strong, Auxiliary outlier data=false, Backbone=ResNet-182024.0595.74——26.9
PALMMethod type=Training, Augmentations=Strong, Auxiliary outlier data=false, Backbone=ResNet-182024.0595.6——25.71
Revisited PCABackbone=ResNetV2-50 (BiT-M)2025.0595.19——17.39
KNNBackbone=ResNet-34 (TIMM)2025.0594.82——31.97
Proj. GradsBackbone=ResNetV2-50 (BiT-M)2025.0594.78——24.13
Softmax scoreBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0894.52——42.31
SSD+Method type=Training, Augmentations=Strong, Auxiliary outlier data=false, Backbone=ResNet-182024.0594.49——35.71
Kernel PCA (CoRP)Backbone=ResNet-34 (TIMM)2025.0594.19——36.9
GradPCABackbone=ResNet-34 (TIMM)2025.0594.19——30.02
Revisited PCABackbone=ResNet-34 (TIMM)2025.0593.29——30.22
GradOrthBackbone=ResNet-34 (TIMM)2025.0593.1——32.6
GENMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0593.05——35.85
EBOMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0592.99——34.99
MaxLogitMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0592.88——36.02
GradPCABackbone=ResNetV2-50 (BiT-M)2025.0592.74——37.65
GradPCA-VecBackbone=ResNet-34 (TIMM)2025.0592.68——44.35
GradPCA-BatchBackbone=ResNet-34 (TIMM)2025.0592.59——44.41
GAIA-ABackbone=ResNetV2-50 (BiT-M)2025.0592.18——35.88
GradPCA+DICEBackbone=ResNetV2-50 (BiT-M)2025.0592.12——41.67
EnergyBackbone=ResNet-34 (TIMM)2025.0592.08——29.51
MSPMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0591.99——49.1
Max logitsBackbone=ResNet-34 (TIMM)2025.0591.98——30.47
ODINBackbone=ResNet-34 (TIMM)2025.0591.98——30.47
MahalanobisBackbone=ResNet-34 (TIMM)2025.0591.83——47.78
GradPCA+DICEBackbone=ResNet-34 (TIMM)2025.0591.7——36.85
GradNormBackbone=DenseNet, In-Distribution Dataset=CIFAR-102023.0891.67——36.89
FedAvg + FOOGDalpha=0.12024.1091.22——37.55
MSPBackbone=ResNet-34 (TIMM)2025.0591.09——47.18
ReActBackbone=ResNet-34 (TIMM)2025.0590.78——31.88
DICEBackbone=ResNet-34 (TIMM)2025.0589.81——32.09
ASHMethod type=Post-hoc, Augmentations=Weak, Auxiliary outlier data=false, Backbone=ResNet-182024.0589.06——42.41
FedRoD + FOOGDalpha=0.12024.1088.69——36.17
Kernel PCA (CoRP)Backbone=ResNetV2-50 (BiT-M)2025.0587.93——58.07
MSPBackbone=ResNetV2-50 (BiT-M)2025.0587.31——37.41
Proj. GradsBackbone=ResNet-34 (TIMM)2025.0585.34——44.21
FedTHEalpha=0.12024.1083.5——43.72
FedICONalpha=0.12024.1082.95——49.98
FedRoDalpha=0.12024.1082.83——43.4
ODINBackbone=ResNetV2-50 (BiT-M)2025.0581.92——45.28
Max logitsBackbone=ResNetV2-50 (BiT-M)2025.0581.91——45.28
GradOrthBackbone=ResNetV2-50 (BiT-M)2025.0581.16——77.5
FedATOLalpha=0.12024.1080.05——61.01
EnergyBackbone=ResNetV2-50 (BiT-M)2025.0579.06——53.08
FedIIRalpha=0.12024.1077.98——57.86
FedAvgalpha=0.12024.1076.29——62.1
FOSTERalpha=0.12024.1076.29——48.73
FedLNalpha=0.12024.1076.03——66.41
GAIA-ZBackbone=ResNetV2-50 (BiT-M)2025.0557.25——92.07
DICEBackbone=ResNetV2-50 (BiT-M)2025.0544——99.88
ReActBackbone=ResNetV2-50 (BiT-M)2025.0539.24——99.41