Out-of-Distribution Detection on PACS Open-Set DTD Food101 Caltech101
99.6DTD AUCCSP
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CSPBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 99.6 | 1.8 | 99.2 | 2.6 | 97.8 | 12.4 | 88.5 | 32.7 | |
| NegLabelBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 99.3 | 4.2 | 97.7 | 15.8 | 95.7 | 31.6 | 86.6 | 49.1 | |
| ODINBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 99.1 | 1.8 | 99.3 | 1 | 97.4 | 6.7 | 92.8 | 17.9 | |
| MCMBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 98.9 | 4.3 | 99.2 | 3.4 | 97 | 13.7 | 93.5 | 27.2 | |
| ∆EnergyBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 98.1 | 6.5 | 99.2 | 2.4 | 96.1 | 14.3 | 93.7 | 25.9 | |
| MSPBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 97.9 | 9.8 | 98.9 | 4.6 | 95.8 | 20.9 | 93.3 | 29.1 | |
| DPLCLIPV-Net=RN502024.05 | 95.4 | 25 | 99 | 5.8 | 97.4 | 14.1 | — | 50.4 | |
| CRoFTV-Net=RN502024.05 | 94.7 | 26.8 | 99.8 | 0.8 | 88.7 | 42 | — | 38.6 | |
| CRoFTV-Net=ViT162024.05 | 94.7 | 33.2 | 99.1 | 6.1 | 88.9 | 52.2 | — | 40 | |
| CLIPNBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 93.6 | 40.3 | 96.2 | 25.6 | 88.1 | 56.1 | 86.6 | 50.9 | |
| DPLCLIPV-Net=ViT162024.05 | 93.2 | 39 | 98 | 13 | 94.4 | 33.5 | — | 61.8 | |
| CoCoOpV-Net=ViT162024.05 | 92.1 | 44.9 | 97.8 | 15 | 87.8 | 57.5 | — | 60.4 | |
| CoCoOpV-Net=RN502024.05 | 90.7 | 42.8 | 98.3 | 9 | 88.4 | 48.5 | — | 58.9 | |
| CoOpV-Net=RN502024.05 | 90.1 | 43.6 | 97 | 15.5 | 83.1 | 56.8 | — | 65.6 | |
| Bayes-CALV-Net=RN502024.05 | 89.7 | 47.8 | 97.4 | 14.8 | 83.5 | 70.5 | — | 66.9 | |
| ReActBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 89.5 | 44.9 | 98.1 | 9.9 | 89.8 | 43.2 | 79.7 | 53.1 | |
| MaxLogitBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 89.5 | 45.1 | 97.9 | 13.1 | 88.9 | 46.8 | 82.6 | 53.3 | |
| CLIP (KNN)V-Net=ViT162024.05 | 87.7 | 58.9 | 98.8 | 5.8 | 82.5 | 70 | — | 52.2 | |
| CLIP-AdapterV-Net=ViT162024.05 | 87.7 | 58.6 | 98.8 | 5.6 | 82.6 | 69.6 | — | 51.9 | |
| Tip-Adapter-FV-Net=ViT162024.05 | 87.3 | 69.3 | 97.8 | 11.2 | 84.8 | 74.9 | — | 64.1 | |
| CoOpV-Net=ViT162024.05 | 85.7 | 62.7 | 96.2 | 22.6 | 80 | 70.9 | — | 66.4 | |
| CLIP (KNN)V-Net=RN502024.05 | 85.3 | 58.5 | 99.1 | 4.3 | 79.2 | 65.3 | — | 49.9 | |
| CLIP-AdapterV-Net=RN502024.05 | 85.3 | 58.2 | 99.1 | 4.2 | 79.2 | 65 | — | 49.4 | |
| Tip-Adapter-FV-Net=RN502024.05 | 83.4 | 81.8 | 97.2 | 15.1 | 84.7 | 72.9 | — | 65.7 | |
| EnergyBackbone=CLIP, Evaluation Protocol=Zero-shot2025.10 | 82.4 | 67.6 | 95.9 | 26 | 86.7 | 52.2 | 76.6 | 61.5 | |
| CLIPV-Net=ViT162024.05 | 82.4 | 67.6 | 95.9 | 26 | 86.7 | 52.2 | — | 61.5 | |
| Bayes-CALV-Net=ViT162024.05 | 79.2 | 62.6 | 87 | 38.6 | 71.8 | 60.3 | — | 69.8 | |
| CLIPV-Net=RN502024.05 | 76.6 | 80.2 | 94.7 | 29.2 | 86.8 | 52.3 | — | 65.8 |