List your company
ResearchFormerly WizWand

Out-of-Distribution Detection on CIFAR10 (ID) vs SVHN (OOD)

100AUROCERD
6.9231.08555.2579.415Dec 10, 2020Nov 6, 2021Oct 3, 2022Aug 31, 2023Jul 27, 2024Jun 23, 2025May 21, 2026
Updated 5mo ago

Evaluation results

MethodLinks
ERDEnsemble size=52020.12100100—
ERD++Ensemble size=52020.12100100—
DynProtoBackbone=CLIP-B/16, Method Category=Vision Language Model based Methods2026.04100—0
AdaNDBackbone=CLIP-B/16, Method Category=Vision Language Model based Methods2026.0499.97—0.15
DynProtoBackbone=ResNet50, Method Category=Vision Model based Methods2026.0499.74—1.49
CIDERContrastive Learning=Yes2026.0199.72—2.89
DALBackbone=ResNet-182024.0599.61—125
KNN+Contrastive Learning=Yes2026.0199.61—2.7
DOEBackbone=ResNet-182024.0599.6—197
Hopfield BoostingBackbone=ResNet-182024.0599.57—23
SSD+Contrastive Learning=Yes2026.0199.51—2.47
POEMBackbone=ResNet-182024.0599.33—148
ViMBackbone=DenseNet-1002026.0399.29—3.39
CE + SimCLRContrastive Learning=Yes2026.0199.22—6.98
MSP-OEBackbone=ResNet-182024.0599.2—431
CADRefBackbone=ResNet50, Method Category=Vision Model based Methods2026.0499.17—4.17
DOSBackbone=ResNet-182024.0599.15—309
EBO-OEBackbone=ResNet-182024.0599.15—266
ViTBackbone=ViT-S-16, Pre-trained=Imagenet21k2020.129998—
Prototype FusionBackbone=ResNet-182026.0398.94—5.92
MCMBackbone=CLIP-B/16, Method Category=Vision Language Model based Methods2026.0498.84—3.6
Prototype FusionBackbone=DenseNet-1002026.0398.74—6.53
ASH-SBackbone=ResNet50, Method Category=Vision Model based Methods2026.0498.74—6.16
DivOEBackbone=ResNet-182024.0598.53—621
VIMBackbone=ResNet50, Method Category=Vision Model based Methods2026.0498.46—8.65
ESOODBackbone=ResNet-182026.0398.19—8.67
LaRExBackbone=DenseNet-1002026.0397.99—10.15
GLMCMBackbone=CLIP-B/16, Method Category=Vision Language Model based Methods2026.0497.94—9.5
ESOODBackbone=DenseNet-1002026.0397.89—11.32
MahalanobisContrastive Learning=No2026.0197.8—9.24
NegRefineBackbone=CLIP-B/16, Method Category=Vision Language Model based Methods2026.0497.57—15.67
MahalanobisBackbone=DenseNet-1002026.0397.18—15.42
EnergyBackbone=DenseNet-1002026.0396.28—14.12
MaxLogitBackbone=DenseNet-1002026.0396.17—14.18
NECOBackbone=DenseNet-1002026.0396.14—14.16
GODINContrastive Learning=No2026.0196.1—18.72
OptFSBackbone=ResNet50, Method Category=Vision Model based Methods2026.0495.97—24.69
ViMBackbone=ResNet-182026.0395.69—22.52
KNNContrastive Learning=No2026.0195.48—27.97
ReActBackbone=DenseNet-1002026.0395.43—17.44
MixOEBackbone=ResNet-182024.0595.37—2,754
LaRExBackbone=ResNet-182026.0395.32—18.5
NECOBackbone=ResNet-182026.0395.04—32.26
DICEBackbone=ResNet50, Method Category=Vision Model based Methods2026.0494.96—27.75
CSIContrastive Learning=Yes2026.0194.69—37.38
SITN2026.0594.6——
NNGuideBackbone=ResNet-182026.0394.59—16.09
ProxyAnchorContrastive Learning=Yes2026.0194.55—39.27
GradNormBackbone=DenseNet-1002026.0394.37—34.82
MaxLogitBackbone=ResNet-182026.0394.36—20.86
EnergyBackbone=ResNet50, Method Category=Vision Model based Methods2026.0494.19—38.79
MahalanobisBackbone=ResNet-182026.0394.12—29
NNGuideBackbone=DenseNet-1002026.0393.78—18.59
CSPBackbone=CLIP-B/16, Method Category=Vision Language Model based Methods2026.0493.73—53.39
EnergyBackbone=ResNet-182026.0393.69—29.2
Comp.VAEContrastive Learning=No2026.0193.6—15.6
MSPBackbone=ResNet-182026.0393.6—19.17
MSPBackbone=ResNet50, Method Category=Vision Model based Methods2026.0493.57—47.23
MSPBackbone=DenseNet-1002026.0393.46—17.71
FedRoD+FOOGDalpha=0.12024.1092.8—17.61
GradNormBackbone=ResNet-182026.0392.76—37.65
ReActBackbone=ResNet-182026.0391.64—46.31
ODINContrastive Learning=No2026.0191.3—53.78
MSPContrastive Learning=No2026.0191.25—59.66
EnergyContrastive Learning=No2026.0191.22—54.41
NegLabelBackbone=CLIP-B/16, Method Category=Vision Language Model based Methods2026.0490.91—48.17
FedRoDalpha=0.12024.1090.16—40.72
FOSTERalpha=0.12024.1085.95—53.24
FedTHEalpha=0.12024.1083.5—48.09
Comp.VAE (vMF)Contrastive Learning=No2026.0181.3—51.8
FedAvg+FOOGDalpha=0.12024.1079.89—44.59
Complexity2026.0578.2——
FedAvgalpha=0.12024.1077.02—66.41
FedATOLalpha=0.12024.1070.95—61.01
KNN*Contrastive Learning=No2026.0166.4—91
FedIIRalpha=0.12024.1066.23—78.34
ReActBackbone=ResNet50, Method Category=Vision Model based Methods2026.0466.05—97.18
FedLNalpha=0.12024.1058.44—80.02
FedICONalpha=0.12024.1039.22—91.75
WAICensemble size=52026.0533.2——
Log-likelihood2026.0510.5——