OOD Detection on iNaturalist
99.71AUROCAdaNeg
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AdaNegLearning Setup=Non-Prompt-Learning Zero-shot CLIP2024.05 | 99.71 | — | — | |
| DynProtoBackbone=CLIP-RN502026.04 | 99.53 | 1.4 | — | |
| NegLabelLearning Setup=Non-Prompt-Learning Zero-shot CLIP2024.05 | 99.49 | — | — | |
| CSPBackbone=CLIP-RN502026.04 | 99.47 | 1.92 | — | |
| NegLabelBackbone=CLIP-RN502026.04 | 99.28 | 2.75 | — | |
| LoCoOp (GL-MCM)Learning Setup=CLIP Prompt Learning 16-shot2024.05 | 99.14 | — | — | |
| CLS-MLearning Setup=CLIP Prompt Learning 16-shot, Base Model=Prompt-SRC2024.05 | 98.68 | — | — | |
| IDLikeBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 98.05 | 9.71 | — | |
| IDLikeBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 97.65 | 12.07 | — | |
| OCOModel=DINOv22026.05 | 97.45 | 10.38 | — | |
| CLS-ELearning Setup=CLIP Prompt Learning 16-shot, Base Model=Prompt-SRC2024.05 | 97.36 | — | — | |
| MaxLogitLearning Setup=CLIP Prompt Learning 16-shot2024.05 | 97.16 | — | — | |
| GL-MCMBackbone=CLIP-ViT-B/16, Setting=Zero-shot2024.11 | 96.71 | 15.16 | — | |
| FDBDModel=DINOv22026.05 | 96.43 | 11.99 | — | |
| VRA++Backbone=BiT, Pre-trained=ImageNet2023.02 | 96.37 | 22.25 | — | |
| LoCoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 96.3 | 17.58 | — | |
| NECOModel=DINOv22026.05 | 96.12 | 12.67 | — | |
| RMDSBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 96.1 | 19.46 | — | |
| MDSBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 96.01 | 20.66 | — | |
| SCTBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 95.86 | 13.94 | — | |
| CoRPModel=DINOv22026.05 | 95.81 | 16.81 | — | |
| MaxLogitModel=DINOv22026.05 | 95.79 | 13.79 | — | |
| SCALEModel=DINOv22026.05 | 95.73 | 13.4 | — | |
| ViMBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 95.72 | 17.59 | — | |
| SCTBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 95.7 | 19.16 | — | |
| EnergyModel=DINOv22026.05 | 95.46 | 14.24 | — | |
| FDBDModel=ViT2026.05 | 95.33 | 18.22 | — | |
| APEXBackbone=ResNet-50, Evaluation Protocol=fine-tuning, Pre-training=pre-trained, In-Distribution (ID) Dataset=ImageNet-1002026.02 | 95.27 | — | 31.32 | |
| APMBackbone=ResNet-50, Evaluation Protocol=fine-tuning, Pre-training=pre-trained, In-Distribution (ID) Dataset=ImageNet-1002026.02 | 95.23 | — | 31.64 | |
| EPDBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 95.17 | 17.84 | — | |
| OpenMaxBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 94.93 | 19.56 | — | |
| PALMBackbone=ResNet-50, Evaluation Protocol=fine-tuning, Pre-training=pre-trained, In-Distribution (ID) Dataset=ImageNet-1002026.02 | 94.82 | — | 32.89 | |
| LoCoOpBackbone=CLIP-RN502026.04 | 94.81 | 27.5 | — | |
| ODIN+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 94.71 | 21.97 | — | |
| MCMLearning Setup=Non-Prompt-Learning Zero-shot CLIP2024.05 | 94.59 | — | — | |
| MCMBackbone=CLIP-ViT-B/16, Setting=Zero-shot2024.11 | 94.17 | 31.86 | — | |
| kNN+Backbone=ResNet-50, Evaluation Protocol=fine-tuning, Pre-training=pre-trained, In-Distribution (ID) Dataset=ImageNet-1002026.02 | 94.15 | — | 38.54 | |
| LoCoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 94.05 | 28.81 | — | |
| EnergyLearning Setup=CLIP Prompt Learning 16-shot2024.05 | 93.97 | — | — | |
| CoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 93.92 | 28.25 | — | |
| OODDModel=DINOv22026.05 | 93.78 | 15.49 | — | |
| MCMBackbone=CLIP-RN502026.04 | 93.72 | 33.9 | — | |
| SHEBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 93.57 | 22.17 | — | |
| GENBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 93.54 | 22.94 | — | |
| SCALEModel=ViT2026.05 | 93.53 | 23.92 | — | |
| CoRPModel=ViT2026.05 | 93.52 | 28.26 | — | |
| SCTBackbone=CLIP-RN502026.04 | 93.38 | 31.14 | — | |
| MaxLogitModel=ViT2026.05 | 93.06 | 26.73 | — | |
| CIDERBackbone=ResNet-50, Evaluation Protocol=fine-tuning, Pre-training=pre-trained, In-Distribution (ID) Dataset=ImageNet-1002026.02 | 92.83 | — | 45.49 | |
| SHEModel=DINOv22026.05 | 92.75 | 31.1 | — | |
| LSNBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 92.66 | 36.17 | — | |
| NECOModel=ViT2026.05 | 92.66 | 33.23 | — | |
| EnergyModel=ViT2026.05 | 92.61 | 27.52 | — | |
| Energy+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 92.41 | 33.03 | — | |
| OCOModel=ViT2026.05 | 92.41 | 30.31 | — | |
| ODIN+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 91.97 | 28.88 | — | |
| MSP+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 91.61 | 36.9 | — | |
| OODDModel=ViT2026.05 | 91.51 | 30.41 | — | |
| KNNBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 91.46 | 27.74 | — | |
| ReActBackbone=BiT, Pre-trained=ImageNet2023.02 | 91.45 | 48.6 | — | |
| CoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 91.4 | 43.8 | — | |
| d(Attractors)feature_space=latent trajectories2025.05 | 91.3 | — | 29.9 | |
| GLMCMBackbone=CLIP-RN502026.04 | 90.94 | 37.28 | — | |
| Energy+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 90.91 | 33.11 | — | |
| MSP+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 90.7 | 37.85 | — | |
| NegPromptBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 90.49 | 37.79 | — | |
| NNguideModel=DINOv22026.05 | 89.96 | 24.83 | — | |
| ODINBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 89.83 | 41.39 | — | |
| ViMBackbone=BiT, Pre-trained=ImageNet2023.02 | 89.3 | 55.71 | — | |
| MSPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 89.06 | 47.83 | — | |
| TempScaleBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 88.54 | 43.08 | — | |
| MSPBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 88.19 | 42.42 | — | |
| MaxLogitBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 88.03 | 60.88 | — | |
| MSPBackbone=BiT, Pre-trained=ImageNet2023.02 | 87.92 | 64.09 | — | |
| EnergyBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 87.69 | 49.12 | — | |
| LSNBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 87.2 | 39.28 | — | |
| EnergyBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 87.18 | 64.98 | — | |
| Mahalanobisfeature_space=features2025.05 | 87 | — | 31 | |
| NNguideModel=ViT2026.05 | 87 | 42.98 | — | |
| ReActBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 86.87 | 65.57 | — | |
| ODINBackbone=BiT, Pre-trained=ImageNet2023.02 | 86.73 | 70.75 | — | |
| ReActBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 86.11 | 48.22 | — | |
| MahalanobisBackbone=BiT, Pre-trained=ImageNet2023.02 | 85.7 | 64.95 | — | |
| MLSBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 85.29 | 72.98 | — | |
| NegPromptBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 84.56 | 65.03 | — | |
| EnergyBackbone=BiT, Pre-trained=ImageNet2023.02 | 84.47 | 74.98 | — | |
| SHEModel=ViT2026.05 | 84.17 | 58.92 | — | |
| DICEBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 82.51 | 47.92 | — | |
| EBOBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 79.3 | 83.58 | — | |
| MSPBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 77.74 | 74.57 | — | |
| KNN2025.05 | 68.6 | — | 86.4 | |
| ODINBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 57.73 | 98.93 | — | |
| ASHBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 50.63 | 97.02 | — | |
| GradNormBackbone=ViT-B/16, ID Accuracy=81.14%, ID Dataset=ImageNet-1K2026.04 | 42.42 | 91.16 | — | |
| Reconstruction2025.05 | 23.6 | — | 99.2 |