List your company
ResearchFormerly WizWand

Out-of-Distribution Detection on SUN OOD with ImageNet-1k In-distribution (test)

98.97AUROCDDE
92.41894.11995.8297.521Nov 18, 2021Aug 24, 2022May 31, 2023Mar 6, 2024Dec 11, 2024Sep 17, 2025Jun 24, 2026
Updated 4mo ago

Evaluation results

MethodLinks
DDEZero-shot=true2026.0698.973.45
Ours (Mean)Training Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.0598.945
ANTSTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.0998.775.43
InterNegArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=no training on ID or extra data2026.0398.686.78
DynProtoExternal Text=false, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0498.557.37
NPOSEvaluation Protocol=fine-tuned2023.0398.489.22
CSP+DCACArchitecture=CLIP-B/162026.0197.5810.18
AdaNegArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=no training on ID or extra data2026.0397.449.5
AdaNegTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.0997.449.5
AdaNegExternal Text=true, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0497.449.5
AdaNeg+NegLabelTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.0597.449.5
AdaNegZero-shot=true2026.0697.449.5
OODDZero-shot=true2026.0696.8416.3
CLIPScopeArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=no training on ID or extra data2026.0396.7715.56
OSPCoOpArchitecture=CLIP-B/162026.0196.7418.26
VOS+Evaluation Protocol=fine-tuned2023.0396.6917.58
CSPArchitecture=CLIP-B/162026.0196.6613.66
CSPArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=no training on ID or extra data2026.0396.6613.66
CSPTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.0996.6613.66
CSPExternal Text=true, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0496.6613.66
CSPTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.0596.6613.66
CMATraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.0996.3616.84
ASH-S+DCACArchitecture=ResNet502026.0196.2117.47
InfoOODTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.0596.1616.06
DynProtoExternal Text=false, Method Category=Vision Model based Methods, Backbone=ResNet-502026.0496.1517.77
LAPTArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=requiring training on ID or extra data2026.0396.0119.12
LAPTTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.0996.0119.12
LAPTTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.0596.0119.12
MCMEvaluation Protocol=zero-shot2023.0395.9525.05
DNMTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.0595.8816.9
AdaNDArchitecture=CLIP-B/162026.0195.8617.08
AdaNDExternal Text=false, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0495.8617.08
AdaNDZero-shot=true2026.0695.8617.08
SUPREMEArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=requiring training on ID or extra data2026.0395.8419.4
SynOODExternal Text=true, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0495.8321.63
SynOODTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.0995.8220.46
GradOrthBackbone=ResNet2023.0895.7619.61
EOEArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=no training on ID or extra data2026.0395.7320.4
EOETraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.0995.7320.4
DDCSArchitecture=ResNet502026.0195.6818.63
NegPromptArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=requiring training on ID or extra data2026.0395.5522.89
NegPromptTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.0595.5522.89
NegLabelArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=no training on ID or extra data2026.0395.4920.53
NegLabelTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.0995.4920.53
NegLabelExternal Text=true, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0495.4920.53
NegLabelTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.0595.4920.53
NegLabelZero-shot=true2026.0695.4920.53
SCTArchitecture=CLIP-B/162026.0195.3320.55
SCTTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.0995.3320.55
HFTTIn-distribution image requirement=Not required2024.0995.2819.24
LINEArchitecture=ResNet502026.0195.2619.48
LINeExternal Text=false, Method Category=Vision Model based Methods, Backbone=ResNet-502026.0495.2619.48
Local-PromptArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=requiring training on ID or extra data2026.0395.1223.23
ASH-BBackbone=ResNet2023.0895.122.08
LoCoOpArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=requiring training on ID or extra data2026.0395.0723.44
LoCoOpExternal Text=false, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0495.0723.44
LoCoOpTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.0595.0723.44
OODDArchitecture=CLIP-B/162026.0195.0121.49
OODDTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.0995.0121.49
EnergyEvaluation Protocol=fine-tuned2023.0394.7718.42
NegRefineExternal Text=true, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0494.6422.93
NegRefineZero-shot=true2026.0694.6322.93
ReActArchitecture=ResNet502026.0194.4124.01
ReActExternal Text=false, Method Category=Vision Model based Methods, Backbone=ResNet-502026.0494.4124.01
LSNArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=requiring training on ID or extra data2026.0394.3526.32
LSNTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.0994.3526.32
LSNTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.0594.3526.32
VRA-PBackbone=ResNet2023.0894.2526.94
ReActBackbone=ResNet-502021.1194.224.2
ReActBackbone=ResNet-50, Pre-trained=ImageNet-1k2021.1194.224.2
ReActBackbone=ResNet2023.0894.224.2
MCMBackbone=CLIP-L, Training Protocol=Zero-shot2022.1194.1429
RankfeatMethods Space=Feature, Backbone=Google BiT-S, Blocks=Block3+42023.1194.0729.27
ASH-S@95Model=VGG-162022.0994.0247.41
ASH-SBackbone=ResNet2023.0894.0227.98
ASH-SArchitecture=ResNet502026.0194.0227.96
ASH-SExternal Text=false, Method Category=Vision Model based Methods, Backbone=ResNet-502026.0494.0227.96
DICE + ReActBackbone=ResNet-50, Pre-trained=ImageNet-1k2021.1193.9425.45
DICE + ReActBackbone=ResNet2023.0893.9425.45
CLIPNArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=requiring training on ID or extra data2026.0393.9326.17
CLIPNTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.0993.9326.17
CLIPNTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.0593.9326.17
EnergyBackbone=CLIP-L, Training Protocol=Fine-tuned2022.1193.8330.46
VRA-DNBackbone=ResNet2023.0893.6530.65
FAExternal Text=false, Method Category=Vision Language Model based Methods, Backbone=CLIP-B/162026.0493.4627.65
CoVerArchitecture=CLIP ViT-B/16, Modality=VLM-based, Training Requirement=no training on ID or extra data2026.0393.4232.85
CoVerTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.0993.4232.85
GL-MCMTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.0593.4129.16
CMAArchitecture=CLIP-B/162026.0193.3229.03
ViMEvaluation Protocol=fine-tuned2023.0393.3131.68
ASH-B@65Model=VGG-162022.0993.1930.04
GradOrthBackbone=MobileNet2023.0893.1830.82
Fort et al./MSPEvaluation Protocol=fine-tuned2023.0393.1638.56
EnergyBackbone=CLIP-B, Training Protocol=Fine-tuned2022.1193.1534.28
OptFSArchitecture=ResNet502026.0193.1335.31
OptFSExternal Text=false, Method Category=Vision Model based Methods, Backbone=ResNet-502026.0493.1335.31
VOSEvaluation Protocol=fine-tuned2023.0392.9129.54
KNNEvaluation Protocol=fine-tuned2023.0392.8726.72
DICE + ReActBackbone=MobileNet2023.0892.8631.22
KNNIn-distribution image requirement=Required2024.0992.6735.62