OOD detection on Places (OOD)
97.31AUROCLoCoOp (GL-MCM)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LoCoOp (GL-MCM)Learning Setup=CLIP Prompt Learning 16-shot2024.05 | 97.31 | — | — | |
| MaxLogitLearning Setup=CLIP Prompt Learning 16-shot2024.05 | 96.88 | — | — | |
| CLS-MLearning Setup=CLIP Prompt Learning 16-shot, Base Model=Prompt-SRC2024.05 | 96.82 | — | — | |
| CLS-ELearning Setup=CLIP Prompt Learning 16-shot, Base Model=Prompt-SRC2024.05 | 96.45 | — | — | |
| EnergyLearning Setup=CLIP Prompt Learning 16-shot2024.05 | 95.74 | — | — | |
| AdaNegLearning Setup=Non-Prompt-Learning Zero-shot CLIP2024.05 | 94.55 | — | — | |
| DynProtoBackbone=CLIP-RN502026.04 | 94.49 | 20.84 | — | |
| LINePost-hoc=true2023.03 | 92.85 | 28.52 | — | |
| ASH-BBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 92.31 | 33.45 | — | |
| SCTBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 92.24 | 29.86 | — | |
| LoCoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 92.03 | 32.21 | — | |
| NegLabelLearning Setup=Non-Prompt-Learning Zero-shot CLIP2024.05 | 91.64 | — | — | |
| LoCoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 91.59 | 33.68 | — | |
| ReActPost-hoc=true2023.03 | 91.58 | 33.85 | — | |
| ReActBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 91.58 | 33.85 | — | |
| SCTBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 91.23 | 32.81 | — | |
| NegPromptBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 91.16 | 35.52 | — | |
| ASH-SBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 90.98 | 39.78 | — | |
| DICE + ReActPost-hoc=true2023.03 | 90.67 | 36.86 | — | |
| DICE + ReActBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 90.67 | 36.86 | — | |
| LSNBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 90.52 | 41.47 | — | |
| CoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 90.5 | 39.12 | — | |
| CoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 90.49 | 40.7 | — | |
| CSPBackbone=CLIP-RN502026.04 | 90.48 | 38.48 | — | |
| GL-MCMBackbone=CLIP-ViT-B/16, Setting=Zero-shot2024.11 | 90.37 | 37.07 | — | |
| MCMLearning Setup=Non-Prompt-Learning Zero-shot CLIP2024.05 | 90.31 | — | — | |
| MCMBackbone=CLIP-ViT-B/16, Setting=Zero-shot2024.11 | 90.09 | 42.94 | — | |
| NegLabelBackbone=CLIP-RN502026.04 | 89.68 | 43.02 | — | |
| LoCoOpBackbone=CLIP-RN502026.04 | 89.59 | 41.53 | — | |
| GradPCABackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 89.02 | 50.86 | — | |
| LSNBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 88.74 | 46.11 | — | |
| IDLikeBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 88.31 | 47.94 | — | |
| KNNBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 88.23 | 69.02 | — | |
| SCTBackbone=CLIP-RN502026.04 | 88.14 | 39.17 | — | |
| IDLikeBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 88.06 | 47.06 | — | |
| DICE + ReActBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 88.02 | 46.78 | — | |
| DICEBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 87.82 | 48.66 | — | |
| ASH-BBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 87.56 | 51.8 | — | |
| DICEPost-hoc=true2023.03 | 87.48 | 46.49 | — | |
| DICEBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 87.48 | 46.49 | — | |
| MaxLogitBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 87.45 | 55.54 | — | |
| ReActBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 87.42 | 56.85 | — | |
| EnergyBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 87.33 | 57.4 | — | |
| ODIN+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 86.99 | 50.06 | — | |
| GLMCMBackbone=CLIP-RN502026.04 | 86.81 | 49.95 | — | |
| ReActBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 86.64 | 51.56 | — | |
| NegPromptBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 86.55 | 51.31 | — | |
| MSP+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 86.07 | 61.94 | — | |
| EnergyBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 85.99 | 57.6 | — | |
| MCMBackbone=CLIP-RN502026.04 | 85.9 | 60.45 | — | |
| ReActBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 85.84 | 56.72 | — | |
| DICEBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 85.81 | 53.11 | — | |
| ASH-PBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 85.58 | 61.79 | — | |
| MSP+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 85.56 | 57.62 | — | |
| Max logitsBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 85.2 | 61.92 | — | |
| ODINBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 85.2 | 61.92 | — | |
| APEXBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 84.91 | 63.97 | — | |
| MSPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 84.89 | 61.76 | — | |
| ASH-SBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 84.73 | 58.84 | — | |
| ODINBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 84.71 | 57.36 | — | |
| APMBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 84.62 | 64.72 | — | |
| ODIN+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 84.12 | 51.22 | — | |
| MahalanobisBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 83.63 | 52.09 | — | |
| ASH-PBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 83.47 | 66.59 | — | |
| Energy+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 83.17 | 61.76 | — | |
| EnergyPost-hoc=true2023.03 | 82.86 | 64.92 | — | |
| Energy scoreBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 82.86 | 64.92 | — | |
| PALMBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 82.62 | 67.8 | — | |
| Energy+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 82.53 | 56.52 | — | |
| ODINBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 82.46 | 60.12 | — | |
| ODINPost-hoc=true2023.03 | 81.78 | 67.89 | — | |
| ODINBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 81.78 | 67.89 | — | |
| MahalanobisBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 81.76 | 67.6 | — | |
| Energy scoreBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 81.19 | 69.37 | — | |
| EnergyBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 80.56 | 66.07 | — | |
| SSD+Backbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 79.9 | 77.74 | — | |
| MSPPost-hoc=true2023.03 | 79.76 | 73.99 | — | |
| Softmax scoreBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 79.76 | 73.99 | — | |
| DMPLBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 79.55 | 73.66 | — | |
| MSPBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 79.32 | 71.23 | — | |
| NPOSBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 78.81 | 67.96 | — | |
| VOSBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 78.63 | 76.85 | — | |
| kNN+Backbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 77.58 | 80.74 | — | |
| ODINBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 76.88 | 87.8 | — | |
| EnergyBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 76.37 | 81.41 | — | |
| Softmax scoreBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 76.27 | 79.23 | — | |
| CSIBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 76.27 | 79.08 | — | |
| ODINBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 75.19 | 82.16 | — | |
| MSPBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 74.21 | 84.38 | — | |
| CIDERBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 73.59 | 79.81 | — | |
| MSPBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 72.18 | 79.72 | — | |
| VimBackbone=ResNet-34, ID Dataset=CIFAR-100, Training Supervision=Labeled2026.02 | 69.34 | 85.34 | — | |
| MahalanobisPost-hoc=true2023.03 | 41.79 | 98.4 | — | |
| MahalanobisBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 41.79 | 98.4 | — | |
| MOSgrouping=taxonomy-based2021.05 | 0.8906 | 0.4954 | 0.9736 | |
| Energy2021.05 | 0.8137 | 0.7302 | 0.9549 | |
| ODIN2021.05 | 0.8067 | 0.7627 | 0.9535 | |
| MSP2021.05 | 0.7676 | 0.8144 | 0.9415 | |
| KL Matching2021.05 | 0.7649 | 0.7261 | 0.9361 | |
| Mahalanobis2021.05 | 0.6446 | 0.8975 | 0.8885 |