Out-of-Distribution Detection on DTD
98AUROCFDBD
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FDBDBackbone=ResNet-502025.01 | 98 | 10.7 | — | |
| SCALEBackbone=ResNet-502025.01 | 97.6 | 11.9 | — | |
| ASHBackbone=ResNet-502025.01 | 97.6 | 11.9 | — | |
| DisCoPatch-64Number of Patches=642025.01 | 96.4 | 18.9 | — | |
| NNGuideBackbone=RegNet2025.01 | 95.8 | 17 | — | |
| RankFeatBackbone=ResNetv2-1012025.01 | 95 | 25.4 | — | |
| MOODv2Backbone=BEiTv22025.01 | 94.3 | 24.7 | — | |
| Ours-DBackbone=SigLIP ViT-B/16, Covariance=Diagonal2025.11 | 94.08 | 24.88 | 96.22 | |
| Ours-DBackbone=CLIP ViT-B/16, Covariance=Diagonal2025.11 | 93 | 33.86 | 91.55 | |
| OursBackbone=CLIP ViT-B/162025.11 | 92.61 | 36.94 | 91.31 | |
| Ours-DBackbone=CLIP ViT-B/32, Covariance=Diagonal2025.11 | 91.56 | 40.23 | 89.44 | |
| OursBackbone=CLIP ViT-B/322025.11 | 91.13 | 46.43 | 89.09 | |
| NNGuideBackbone=ResNet-502025.01 | 90.4 | 27.4 | — | |
| OursBackbone=SigLIP ViT-B/162025.11 | 89.44 | 68.58 | 94.27 | |
| NNGuideBackbone=MobileNetV22025.01 | 88.3 | 40.6 | — | |
| TrustVLM*Backbone=SigLIP ViT-B/16, Extra features=DinoV22025.11 | 82.84 | 79.9 | 92.78 | |
| PCME++Backbone=CLIP ViT-B/322025.11 | 82.06 | 75.65 | 93.15 | |
| MaxSoftmaxBackbone=SigLIP ViT-B/162025.11 | 82.05 | 82.29 | 89.21 | |
| TempScalingBackbone=SigLIP ViT-B/162025.11 | 81.48 | 72.41 | 88.84 | |
| EntropyBackbone=SigLIP ViT-B/162025.11 | 80.3 | 71.18 | 88.18 | |
| BayesVLMBackbone=SigLIP ViT-B/162025.11 | 80.06 | 61.11 | 78.4 | |
| MaxSoftmaxBackbone=CLIP ViT-B/162025.11 | 79.96 | 76.33 | 78.54 | |
| TempScalingBackbone=CLIP ViT-B/162025.11 | 79.94 | 76.33 | 78.48 | |
| EntropyBackbone=CLIP ViT-B/162025.11 | 79.15 | 76.86 | 77.54 | |
| ProLIPBackbone=CLIP ViT-B/162025.11 | 78.96 | 72.74 | 86.7 | |
| TrustVLM*Backbone=CLIP ViT-B/16, Extra features=DinoV22025.11 | 77.17 | 74.49 | 88.98 | |
| TempScalingBackbone=CLIP ViT-B/322025.11 | 77.01 | 77.56 | 74.36 | |
| MaxSoftmaxBackbone=CLIP ViT-B/322025.11 | 76.98 | 78.49 | 74.32 | |
| ZeroBackbone=CLIP ViT-B/162025.11 | 76.64 | 81.79 | 77.68 | |
| EntropyBackbone=CLIP ViT-B/322025.11 | 76.43 | 80.14 | 73.92 | |
| ZeroBackbone=SigLIP ViT-B/162025.11 | 75.65 | 74.22 | 86.51 | |
| BayesVLMBackbone=CLIP ViT-B/322025.11 | 73.77 | 77.32 | 58.74 | |
| MaxCosineBackbone=CLIP ViT-B/162025.11 | 69.69 | 85.67 | 64.96 | |
| MaxCosineBackbone=SigLIP ViT-B/162025.11 | 68.82 | 85.19 | 78.99 | |
| MaxCosineBackbone=CLIP ViT-B/322025.11 | 66.16 | 86.56 | 58.61 | |
| ProbVLMBackbone=CLIP ViT-B/322025.11 | 46.57 | 97.26 | 39.52 |