OOD Detection on SUN (OOD)
95.33AUROCSCT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SCTBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 95.33 | 20.55 | — | |
| LoCoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 95.2 | 22.82 | — | |
| ASH-BBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 95.1 | 22.08 | — | |
| VRA+Backbone=ResNet-50, Pre-trained=ImageNet2023.02 | 94.91 | 23.5 | — | |
| SCTBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 94.58 | 23.52 | — | |
| LoCoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 94.51 | 25.76 | — | |
| VRABackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 94.25 | 26.94 | — | |
| ReActBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 94.2 | 24.2 | — | |
| ReActBackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 94.2 | 24.2 | — | |
| ASH-SBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 94.02 | 27.98 | — | |
| DICE + ReActBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 93.94 | 25.45 | — | |
| LSNBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 93.53 | 34.27 | — | |
| GL-MCMBackbone=CLIP-ViT-B/16, Setting=Zero-shot2024.11 | 93.41 | 29.16 | — | |
| VRABackbone=ResNetv2-101, Pre-trained=ImageNet2023.02 | 93.27 | 34.53 | — | |
| CoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 93.13 | 31.15 | — | |
| DICE + ReActBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 92.86 | 31.22 | — | |
| VRA+Backbone=ResNetv2-101, Pre-trained=ImageNet2023.02 | 92.68 | 32.89 | — | |
| CoOpBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 92.65 | 35.42 | — | |
| MCMBackbone=CLIP-ViT-B/16, Setting=Zero-shot2024.11 | 92.55 | 37.28 | — | |
| NegPromptBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 92.25 | 32.11 | — | |
| MOSgrouping=taxonomy-based2021.05 | 92.01 | 40.63 | 98.17 | |
| MOSBackbone=ResNetv2-101, Pre-trained=ImageNet2023.02 | 92.01 | 40.63 | — | |
| ASH-BBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 91.61 | 38.45 | — | |
| LSNBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 91.47 | 40.15 | — | |
| GradPCABackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 91.27 | 42.8 | — | |
| EnergyBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 91.17 | 46.42 | — | |
| MaxLogitBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 91.16 | 44.83 | — | |
| IDLikeBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 91.07 | 40.55 | — | |
| ReActBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 91.04 | 46.17 | — | |
| DICEBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 90.83 | 35.15 | — | |
| DICEBackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 90.83 | 35.15 | — | |
| DICEBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 90.69 | 41.81 | — | |
| IDLikeBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=16-shot2024.11 | 90.54 | 38.93 | — | |
| DICEBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 90.46 | 38.69 | — | |
| ASH-SBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 90.02 | 43.62 | — | |
| NegPromptBackbone=CLIP-ViT-B/16, Setting=Prompt tuning, Shots=1-shot2024.11 | 89.63 | 44.39 | — | |
| ReActBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 89.28 | 47.86 | — | |
| EnergyBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 89.18 | 49.73 | — | |
| ASH-PBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 88.35 | 52.88 | — | |
| ReActBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 88.16 | 47.69 | — | |
| Max logitsBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 88.12 | 55.32 | — | |
| ODINBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 88.12 | 55.32 | — | |
| ReActBackbone=ResNetv2-101, Pre-trained=ImageNet2023.02 | 87.4 | 65.3 | — | |
| ODIN+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 86.92 | 49.69 | — | |
| ASH-PBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 86.72 | 58.61 | — | |
| MahalanobisBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 86.33 | 47.82 | — | |
| KNNBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 86.06 | 62.08 | — | |
| Energy scoreBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 85.89 | 59.26 | — | |
| EnergyBackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 85.89 | 59.26 | — | |
| ODINBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 85.88 | 54.07 | — | |
| MahalanobisBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 85.68 | 61.12 | — | |
| MSP+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 85.51 | 57.36 | — | |
| Energy2021.05 | 85.32 | 65.33 | 96.57 | |
| EnergyBackbone=ResNetv2-101, Pre-trained=ImageNet2023.02 | 85.32 | 65.33 | — | |
| MSPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 85.02 | 61.36 | — | |
| MSP+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 84.91 | 61.56 | — | |
| SHEBackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 84.69 | 54.19 | — | |
| ODINBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 84.59 | 60.15 | — | |
| ODINBackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 84.59 | 60.15 | — | |
| Energy scoreBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 84.5 | 62.65 | — | |
| ODIN+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 84.01 | 51.06 | — | |
| ODIN2021.05 | 83.92 | 71.67 | 96.26 | |
| ODINBackbone=ResNetv2-101, Pre-trained=ImageNet2023.02 | 83.92 | 71.67 | — | |
| Energy+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 83.26 | 60.99 | — | |
| ODINBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 82.66 | 58.52 | — | |
| Energy+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 82.64 | 55.44 | — | |
| EnergyBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 80.96 | 65 | — | |
| MSPBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 80.92 | 66.99 | — | |
| Softmax scoreBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 80.86 | 70.83 | — | |
| MSPBackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 80.86 | 70.83 | — | |
| KNNBackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 80.1 | 69.53 | — | |
| KL Matching2021.05 | 78.72 | 67.52 | 94.1 | |
| ODINBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 78.42 | 88.72 | — | |
| MSP2021.05 | 78.34 | 79.98 | 94.45 | |
| MSPBackbone=ResNetv2-101, Pre-trained=ImageNet2023.02 | 78.34 | 79.98 | — | |
| Softmax scoreBackbone=MobileNet, In-Distribution Dataset=ImageNet-1k2022.09 | 77.1 | 77.02 | — | |
| MSPBackbone=CLIP-ViT-B/16, Setting=Post-hoc2024.11 | 73.97 | 76.95 | — | |
| Mahalanobis2021.05 | 65.2 | 88.43 | 88.81 | |
| MahalanobisBackbone=ResNetv2-101, Pre-trained=ImageNet2023.02 | 65.2 | 88.43 | — | |
| MahalanobisBackbone=ResNet, In-Distribution Dataset=ImageNet-1k2022.09 | 42.41 | 98.5 | — | |
| MahalanobisBackbone=ResNet-50, Pre-trained=ImageNet2023.02 | 42.41 | 98.5 | — |