Out-of-Distribution Detection on VLCS Open-Set (DTD, Food101, Caltech101)
88.9AUC (DTD)CSP
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CSPBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 88.9 | 38.7 | 78.3 | 67.4 | 67.3 | 73 | 88.5 | 32.7 | |
| CRoFTV-Net=ViT162024.05 | 86.7 | 60.6 | 96 | 25.9 | 83.7 | 62.2 | — | 40 | |
| CRoFTV-Net=RN502024.05 | 86.6 | 58.8 | 93.9 | 35.3 | 77.7 | 68.1 | — | 38.6 | |
| ∆EnergyBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 85.3 | 53.2 | 94.1 | 31.9 | 89.5 | 47.3 | 93.7 | 25.9 | |
| NegLabelBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 84.8 | 54.3 | 79.4 | 74.6 | 62.7 | 84.3 | 86.6 | 49.1 | |
| MSPBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 84.4 | 55.3 | 93.7 | 35.1 | 88.9 | 48.8 | 93.3 | 29.1 | |
| MCMBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 84.2 | 55.1 | 93.3 | 36.4 | 88.5 | 50.4 | 93.5 | 27.2 | |
| ODINBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 83.1 | 46.5 | 91.3 | 28.2 | 86.8 | 40.9 | 92.8 | 17.9 | |
| CLIP-AdapterV-Net=ViT162024.05 | 82.6 | 71.2 | 95.1 | 32.4 | 78.8 | 74.2 | — | 51.9 | |
| CLIP (KNN)V-Net=RN502024.05 | 82.4 | 67.4 | 95.2 | 27 | 74.1 | 74.4 | — | 49.9 | |
| CLIP (KNN)V-Net=ViT162024.05 | 82.4 | 71.6 | 95.1 | 32.6 | 78.7 | 74.1 | — | 52.2 | |
| CLIP-AdapterV-Net=RN502024.05 | 82.4 | 67.7 | 95.3 | 26.5 | 74.1 | 74.7 | — | 49.4 | |
| CLIPNBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 80.4 | 62.7 | 88.9 | 46.6 | 72.5 | 74.1 | 86.6 | 50.9 | |
| Tip-Adapter-FV-Net=ViT162024.05 | 66.7 | 94.4 | 92.7 | 39.6 | 63 | 95.1 | — | 64.1 | |
| MaxLogitBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 60.8 | 87.6 | 88.9 | 45.4 | 69.4 | 81.6 | 82.6 | 53.3 | |
| Tip-Adapter-FV-Net=RN502024.05 | 59.3 | 95.4 | 91.5 | 38.6 | 64.1 | 84 | — | 65.7 | |
| DPLCLIPV-Net=RN502024.05 | 56.3 | 93.8 | 79.9 | 67.6 | 51.5 | 96 | — | 50.4 | |
| EnergyBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 55.3 | 88.8 | 85.8 | 48.3 | 53.3 | 86.3 | 76.6 | 61.5 | |
| CLIPV-Net=ViT162024.05 | 55.3 | 88.8 | 85.8 | 48.3 | 53.3 | 86.3 | — | 61.5 | |
| DPLCLIPV-Net=ViT162024.05 | 52.9 | 98.7 | 69.7 | 88.8 | 52.4 | 97.3 | — | 61.8 | |
| ReActBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 52.8 | 89.7 | 86.7 | 47.8 | 61.4 | 83.2 | 79.7 | 53.1 | |
| Bayes-CALV-Net=RN502024.05 | 50.9 | 94.3 | 69.5 | 76 | 53.4 | 97.7 | — | 66.9 | |
| CoCoOpV-Net=ViT162024.05 | 49.2 | 93.3 | 78.4 | 60.8 | 49.2 | 90.6 | — | 60.4 | |
| CoOpV-Net=ViT162024.05 | 47.3 | 89.2 | 76.3 | 62.9 | 46.8 | 90 | — | 66.4 | |
| Bayes-CALV-Net=ViT162024.05 | 44.2 | 93.6 | 68.1 | 72.9 | 34 | 90.7 | — | 69.8 | |
| CoCoOpV-Net=RN502024.05 | 42.4 | 95.9 | 80.2 | 65 | 46.8 | 92.5 | — | 58.9 | |
| CLIPV-Net=RN502024.05 | 40.6 | 94.5 | 82.8 | 52.8 | 61.5 | 85.5 | — | 65.8 | |
| CoOpV-Net=RN502024.05 | 40.4 | 96.9 | 54.5 | 87 | 46.3 | 93.7 | — | 65.6 |