OOD Detection on ImageNet-1k ID Average OOD
0.9797AUROCTANL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TANLEncoder=VITB/16 CLIP, Protocol=Training-free & test-time adaptation methods2026.03 | 0.9797 | 0.0981 | |
| ∆Energy2025.10 | 0.9737 | 0.0961 | |
| TTLExtra Resources=No, Category=Test-time adaptation2026.04 | 0.9729 | 0.1246 | |
| MaxLogit2025.10 | 0.9706 | 0.1155 | |
| Energy2025.10 | 0.9702 | 0.1168 | |
| SynOODEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9701 | 0.1427 | |
| ReAct2025.10 | 0.9677 | 0.1216 | |
| AdaNegBackbone=ViT-B/16 CLIP, Training Status=Zero-Shot Training-free2024.10 | 0.9666 | 0.1892 | |
| AdaNegEncoder=VITB/16 CLIP, Protocol=Training-free & test-time adaptation methods2026.03 | 0.9666 | 0.1892 | |
| CLIPN2025.10 | 0.965 | 0.1353 | |
| ODIN2025.10 | 0.9642 | 0.1492 | |
| AdaNegExtra Resources=Yes, Category=Test-time adaptation2026.04 | 0.9617 | 0.1922 | |
| Peng et al.Encoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 0.96 | 0.2236 | |
| OODDEncoder=VITB/16 CLIP, Protocol=Training-free & test-time adaptation methods2026.03 | 0.96 | 0.1879 | |
| CSP+∆EnergyEvaluation Protocol=Zero-shot methods2025.10 | 0.9583 | 0.1622 | |
| CSPEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 0.9576 | 0.1751 | |
| CSPExtra Resources=Yes, Category=Post-hoc2026.04 | 0.9576 | 0.1751 | |
| CSPEvaluation Protocol=Zero-shot methods2025.10 | 0.9576 | 0.1751 | |
| AdaNDEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9558 | 0.16 | |
| CLIPScopeEncoder=VITB/16 CLIP, Protocol=Training-free & test-time adaptation methods2026.03 | 0.953 | 0.2088 | |
| CMA [30]Encoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9513 | 0.1993 | |
| MoFEExtra Resources=No, Category=Training-based2026.04 | 0.9489 | 0.2002 | |
| NegPromptBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.9481 | 0.2301 | |
| NegPromptEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9481 | 0.2301 | |
| NegPromptEvaluation Protocol=Tuning-based methods2025.10 | 0.9481 | 0.2301 | |
| OSPCoOpExtra Resources=Yes, Category=Training-based2026.04 | 0.9475 | 0.2513 | |
| DualCnstEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 0.9472 | 0.2145 | |
| LAPTBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.9468 | 0.234 | |
| LAPTEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9468 | 0.234 | |
| NegRefineExtra Resources=Yes, Category=Post-hoc2026.04 | 0.9458 | 0.2201 | |
| NegLabelBackbone=ViT-B/16 CLIP, Training Status=Zero-Shot Training-free2024.10 | 0.9421 | 0.254 | |
| NegLabelEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 0.9421 | 0.254 | |
| NegLabelEvaluation Protocol=Zero-shot methods2025.10 | 0.9421 | 0.254 | |
| NeglabelExtra Resources=Yes, Category=Post-hoc2026.04 | 0.9413 | 0.261 | |
| OODDExtra Resources=No, Category=Test-time adaptation2026.04 | 0.9409 | 0.2364 | |
| AdaNDExtra Resources=No, Category=Test-time adaptation2026.04 | 0.94 | 0.245 | |
| ID-LikeEvaluation Protocol=Tuning-based methods2025.10 | 0.9383 | 0.2951 | |
| FAGLEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9382 | 0.2568 | |
| FAExtra Resources=No, Category=Training-based2026.04 | 0.9378 | 0.2575 | |
| MCM2025.10 | 0.9377 | 0.2583 | |
| Local-PromptExtra Resources=No, Category=Training-based2026.04 | 0.9365 | 0.2832 | |
| APTExtra Resources=Yes, Category=Training-based2026.04 | 0.9364 | 0.2603 | |
| LoCoOpBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.9352 | 0.2866 | |
| LoCoOpEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9352 | 0.2866 | |
| LoCoOpExtra Resources=No, Category=Training-based2026.04 | 0.9352 | 0.2866 | |
| LoCoOpEvaluation Protocol=Tuning-based methods2025.10 | 0.9352 | 0.2866 | |
| GalLoPEvaluation Protocol=Tuning-based methods2025.10 | 0.932 | 0.273 | |
| CLIPNExtra Resources=Yes, Category=Post-hoc2026.04 | 0.9319 | 0.3021 | |
| CLIPNBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.931 | 0.311 | |
| CLIPNEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.931 | 0.311 | |
| CLIPNEvaluation Protocol=Zero-shot methods2025.10 | 0.931 | 0.311 | |
| CMA [37]Encoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 0.9305 | 0.3279 | |
| CMAExtra Resources=Yes, Category=Post-hoc2026.04 | 0.9305 | 0.3279 | |
| LSNBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.9296 | 0.3022 | |
| LSNEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9296 | 0.3022 | |
| EOEEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 0.9296 | 0.3009 | |
| LSN+CoCoOpEvaluation Protocol=Tuning-based methods2025.10 | 0.9296 | 0.3022 | |
| CoVerEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 0.9245 | 0.3488 | |
| TagOODEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.924 | 0.2785 | |
| LSN+CoOpEvaluation Protocol=Tuning-based methods2025.10 | 0.9233 | 0.3197 | |
| EnergyBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.9154 | 0.3989 | |
| EnergyEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9154 | 0.3989 | |
| NPOSBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.9122 | 0.3793 | |
| NPOSEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9122 | 0.3793 | |
| VOSBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.9119 | 0.4132 | |
| VOSEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9119 | 0.4132 | |
| KNNBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.9097 | 0.4219 | |
| KNNEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.9097 | 0.4219 | |
| GL-MCMExtra Resources=No, Category=Post-hoc2026.04 | 0.909 | 0.3504 | |
| MCMBackbone=ViT-B/16 CLIP, Training Status=Zero-Shot Training-free2024.10 | 0.9082 | 0.4393 | |
| MCMEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 0.9082 | 0.4393 | |
| MCMExtra Resources=No, Category=Post-hoc2026.04 | 0.9076 | 0.4277 | |
| MCMEvaluation Protocol=Zero-shot methods2025.10 | 0.9076 | 0.4274 | |
| IDLikeExtra Resources=No, Category=Training-based2026.04 | 0.9052 | 0.4117 | |
| MaxLogitEvaluation Protocol=CLIP-based post-hoc methods2025.10 | 0.8882 | 0.5249 | |
| ODINBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.888 | 0.4775 | |
| ODINEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.888 | 0.4775 | |
| ODINEvaluation Protocol=CLIP-based post-hoc methods2025.10 | 0.888 | 0.4775 | |
| EnergyEvaluation Protocol=CLIP-based post-hoc methods2025.10 | 0.8848 | 0.548 | |
| ReActEvaluation Protocol=CLIP-based post-hoc methods2025.10 | 0.8837 | 0.5462 | |
| ViMBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.8782 | 0.502 | |
| ViMEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.8782 | 0.502 | |
| ZOCBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.8179 | 0.8519 | |
| ZOCEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.8179 | 0.8519 | |
| MSPBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.8163 | 0.6961 | |
| MSPEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.8163 | 0.6961 | |
| EnergyBackbone=ViT-B/16 CLIP, Training Status=Zero-Shot Training-free2024.10 | 0.7957 | 0.8221 | |
| MSP2025.10 | 0.7702 | 0.7083 | |
| MSPEvaluation Protocol=CLIP-based post-hoc methods2025.10 | 0.7498 | 0.7622 | |
| GradNormBackbone=ViT-B/16 CLIP, Training Status=Methods requiring training2024.10 | 0.7235 | 0.8082 | |
| GradNormEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 0.7235 | 0.8082 | |
| MahalanobisBackbone=ViT-B/16 CLIP, Training Status=Zero-Shot Training-free2024.10 | 0.615 | 0.9894 |