Out-of-Distribution Detection on ImageNet-1k (ID) vs Textures (OOD)
98.48AUROCPrototype Fusion
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Prototype FusionBackbone=RegNet2026.03 | 98.48 | — | 7.2 | |
| LareXBackbone=RegNet2026.03 | 97.76 | — | 15.61 | |
| ANTSTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.09 | 96.38 | 18.52 | — | |
| LareXBackbone=ResNet-502026.03 | 96.34 | — | 20.49 | |
| Prototype FusionBackbone=ResNet-502026.03 | 96.08 | — | 19.55 | |
| SynOODTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 95.29 | 22.94 | — | |
| AdaNegTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.09 | 94.93 | 31.27 | — | |
| ID-LikeTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 94.32 | 25.27 | — | |
| CSPTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.09 | 93.86 | 25.52 | — | |
| OODDTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.09 | 93.27 | 30.69 | — | |
| LoCoOpTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 93.24 | 33.06 | — | |
| ReActBackbone=RegNet2026.03 | 91.69 | — | 35.21 | |
| CMATraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 91.64 | 33.58 | — | |
| LAPTTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 91.06 | 40.32 | — | |
| CLIPNTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 90.93 | 40.83 | — | |
| LSNTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 90.42 | 38.54 | — | |
| NegLabelTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.09 | 90.22 | 43.56 | — | |
| CoVerTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.09 | 90.14 | 43.39 | — | |
| ViMBackbone=RegNet2026.03 | 89.62 | — | 44.16 | |
| NNGuideBackbone=RegNet2026.03 | 89.6 | — | 35.76 | |
| SCTTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 89.06 | 41.51 | — | |
| NegPromptTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 88.38 | 43.93 | — | |
| MahalanobisBackbone=ResNet-502026.03 | 87.13 | — | 53.09 | |
| MCMTraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.09 | 86.12 | 58.5 | — | |
| EOETraining Protocol=Zero Shot (No Training Required), Encoder=ViT-B/162025.09 | 85.64 | 57.63 | — | |
| MahalanobisBackbone=RegNet2026.03 | 84.98 | — | 53.62 | |
| NECOBackbone=RegNet2026.03 | 84.19 | — | 47.98 | |
| GradNormBackbone=RegNet2026.03 | 81.86 | — | 67.5 | |
| EnergyBackbone=RegNet2026.03 | 81.46 | — | 64.85 | |
| MSPBackbone=RegNet2026.03 | 80.74 | — | 64.79 | |
| ESOODBackbone=RegNet2026.03 | 80.23 | — | 40.65 | |
| MSPTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 79.69 | 71.93 | — | |
| ZOCTraining Protocol=Training-required (or with Fine-tuning), Encoder=ViT-B/162025.09 | 76.46 | 98.9 | — | |
| MSPBackbone=ResNet-502026.03 | 74.78 | — | 87.76 | |
| MaxLogitBackbone=RegNet2026.03 | 73.44 | — | 69.48 | |
| NNGuideBackbone=ResNet-502026.03 | 72.39 | — | 71.51 | |
| MaxLogitBackbone=ResNet-502026.03 | 69.25 | — | 91.74 | |
| ESOODBackbone=ResNet-502026.03 | 68.87 | — | 76.89 | |
| NECOBackbone=ResNet-502026.03 | 51.02 | — | 89.28 | |
| ViMBackbone=ResNet-502026.03 | 49.41 | — | 93.88 | |
| EnergyBackbone=ResNet-502026.03 | 48.57 | — | 94.28 | |
| GradNormBackbone=ResNet-502026.03 | 39.25 | — | 91.06 | |
| ReActBackbone=ResNet-502026.03 | 35.14 | — | 94.52 |