Out-of-Distribution Detection on ImageNet-1k (ID) vs Textures (OOD) 1.0 (test)
98.93AUCZODE-KNN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ZODE-KNN2022.12 | 98.93 | 4.57 | — | |
| TANLEncoder=VITB/16 CLIP, Protocol=Training-free & test-time adaptation methods2026.03 | 97.11 | — | 13.38 | |
| KNN2022.12 | 97.07 | 11.77 | — | |
| Ours (Mean)Training Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 97.01 | — | 17.12 | |
| KNN+2022.12 | 95.4 | 15.55 | — | |
| SynOODEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 95.29 | — | 22.94 | |
| AdaNegEncoder=VITB/16 CLIP, Protocol=Training-free & test-time adaptation methods2026.03 | 94.93 | — | 31.27 | |
| AdaNeg+NegLabelTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 94.93 | — | 31.27 | |
| OODDEncoder=VITB/16 CLIP, Protocol=Training-free & test-time adaptation methods2026.03 | 94.51 | — | 30.67 | |
| CSPEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 93.86 | — | 25.52 | |
| CSPTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 93.86 | — | 25.52 | |
| AdaNDEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 93.01 | — | 21.76 | |
| FAGLEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 92.93 | — | 29.5 | |
| Peng et al.Encoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 92.15 | — | 38.79 | |
| DualCnstEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 91.73 | — | 34.91 | |
| CMA [30]Encoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 91.64 | — | 33.58 | |
| NegPromptEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 91.6 | — | 35.21 | |
| NegPromptTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 91.6 | — | 35.21 | |
| CLIPScopeEncoder=VITB/16 CLIP, Protocol=Training-free & test-time adaptation methods2026.03 | 91.41 | — | 38.37 | |
| LAPTEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 91.06 | — | 40.32 | |
| LAPTTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 91.06 | — | 40.32 | |
| InfoOODTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 91.01 | — | 40.78 | |
| CLIPNEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 90.93 | — | 40.83 | |
| CLIPNTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 90.93 | — | 40.83 | |
| DNMTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 90.72 | — | 38.67 | |
| TagOODEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 90.6 | — | 36.31 | |
| LSNEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 90.42 | — | 38.54 | |
| LSNTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 90.42 | — | 38.54 | |
| NegLabelEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 90.22 | — | 43.56 | |
| NegLabelTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 90.22 | — | 43.56 | |
| LoCoOpEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 90.19 | — | 42.28 | |
| LoCoOpTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 90.19 | — | 42.28 | |
| CoVerEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 90.14 | — | 43.39 | |
| NPOSEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 88.8 | — | 46.12 | |
| NPOSTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 88.8 | — | 46.12 | |
| SSD+2022.12 | 88.52 | 36.37 | — | |
| CMA [37]Encoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 88.47 | — | 47.35 | |
| ODINEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 87.85 | — | 51.67 | |
| ODINTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 87.85 | — | 51.67 | |
| ViMEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 87.18 | — | 53.94 | |
| ViMTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 87.18 | — | 53.94 | |
| EnergyEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 86.76 | — | 57.61 | |
| EnergyTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 86.76 | — | 57.61 | |
| VOSEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 86.33 | — | 61.02 | |
| VOSTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 86.33 | — | 61.02 | |
| MCMEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 86.12 | — | 58.5 | |
| MCMTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 86.12 | — | 58.5 | |
| Energy2022.12 | 85.99 | 53.72 | — | |
| KNNEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 85.67 | — | 64.35 | |
| KNNTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 85.67 | — | 64.35 | |
| EOEEncoder=VITB/16 CLIP, Protocol=Training-free & non-adaptive methods2026.03 | 85.64 | — | 57.53 | |
| ODIN2022.12 | 85.62 | 50.23 | — | |
| Mahalanobis2022.12 | 85.01 | 55.8 | — | |
| GL-MCMTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 83.11 | — | 55.85 | |
| MSPEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 79.69 | — | 71.93 | |
| MSPTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 79.69 | — | 71.93 | |
| MSP2022.12 | 79.61 | 68 | — | |
| ZOCEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 76.46 | — | 98.9 | |
| ZOCTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 76.46 | — | 98.9 | |
| GODIN2022.12 | 73.27 | 77.85 | — | |
| GradNormEncoder=VITB/16 CLIP, Protocol=Methods requiring training (or fine-tuning)2026.03 | 70.26 | — | 79.36 | |
| GradNormTraining Protocol=Training/Fine-tuning, Encoder=ViT B/16 CLIP2026.05 | 70.26 | — | 79.36 | |
| EnergyTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 65.56 | — | 93.65 | |
| MahalanobisTraining Protocol=Zero-shot Training-free, Encoder=ViT B/16 CLIP2026.05 | 64.23 | — | 98.46 |