Out-of-Distribution Detection on ImageNet-1k OOD Average (val)
22.37FPR95Debiased Negative Mining
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Debiased Negative MiningTraining Requirement=Zero-Shot Training-free Methods, Backbone=ViT B/16 CLIP encoder2026.05 | 22.37 | 94.52 | |
| NegPromptTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 23.01 | 94.81 | |
| LAPTTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 23.4 | 94.68 | |
| NegLabelTraining Requirement=Zero-Shot Training-free Methods, Backbone=ViT B/16 CLIP encoder2026.05 | 25.4 | 94.21 | |
| LoCoOpTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 28.66 | 93.52 | |
| LSNTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 30.22 | 92.96 | |
| CLIPNTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 31.1 | 93.1 | |
| NPOSTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 37.93 | 91.22 | |
| EnergyTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 39.89 | 91.54 | |
| Energy Profile Divergence (EPD)Backbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 40.96 | 87.26 | |
| VOSTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 41.32 | 91.19 | |
| KNNTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 42.19 | 90.97 | |
| RMDSBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 43.4 | 87.6 | |
| MCMTraining Requirement=Zero-Shot Training-free Methods, Backbone=ViT B/16 CLIP encoder2026.05 | 43.93 | 90.82 | |
| MDSBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 44.43 | 87.18 | |
| SHEBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 44.63 | 85.9 | |
| ViMBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 47.02 | 86.51 | |
| KNNBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 47.35 | 84.13 | |
| GENBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 47.65 | 85.33 | |
| ODINTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 47.75 | 88.8 | |
| ViMTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 50.2 | 87.82 | |
| MSPBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 63.75 | 81.03 | |
| TempScaleBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 66.11 | 81.06 | |
| ReActBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 66.15 | 79.12 | |
| OpenMaxBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 68.81 | 83.02 | |
| MSPTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 69.61 | 81.63 | |
| GradNormTraining Requirement=Methods requiring training (or fine-tuning), Backbone=ViT B/16 CLIP encoder2026.05 | 80.82 | 72.35 | |
| EnergyTraining Requirement=Zero-Shot Training-free Methods, Backbone=ViT B/16 CLIP encoder2026.05 | 82.21 | 79.57 | |
| ZOCTraining Requirement=Zero-Shot Training-free Methods, Backbone=ViT B/16 CLIP encoder2026.05 | 85.19 | 81.79 | |
| GradNormBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 93.47 | 46.52 | |
| ASHBackbone=ViT-B/16, ID Accuracy=81.14%2026.04 | 95.84 | 52.22 | |
| MahalanobisTraining Requirement=Zero-Shot Training-free Methods, Backbone=ViT B/16 CLIP encoder2026.05 | 98.94 | 61.5 |