Error Detection on ImageNet
47FDRRTS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RTSBudget size=b=12%, Backbone=ResNet1012026.05 | 47 | — | — | — | |
| MarginBudget size=b=3%, Backbone=ResNet1012026.05 | 48 | — | — | — | |
| MarginBudget size=b=10%, Backbone=ResNet1012026.05 | 48 | — | — | — | |
| MarginBudget size=b=12%, Backbone=ResNet1012026.05 | 48 | — | — | — | |
| RTSBudget size=b=10%, Backbone=ResNet1012026.05 | 49 | — | — | — | |
| MarginBudget size=b=5%, Backbone=ResNet1012026.05 | 51 | — | — | — | |
| MarginBudget size=b=7%, Backbone=ResNet1012026.05 | 51 | — | — | — | |
| DeepGiniBudget size=b=12%, Backbone=ResNet1012026.05 | 51 | — | — | — | |
| VanillaBudget size=b=12%, Backbone=ResNet1012026.05 | 51 | — | — | — | |
| DATISBudget size=b=12%, Backbone=ResNet1012026.05 | 51 | — | — | — | |
| RTSBudget size=b=7%, Backbone=ResNet1012026.05 | 52 | — | — | — | |
| VanillaBudget size=b=10%, Backbone=ResNet1012026.05 | 52 | — | — | — | |
| DATISBudget size=b=10%, Backbone=ResNet1012026.05 | 52 | — | — | — | |
| DeepGiniBudget size=b=10%, Backbone=ResNet1012026.05 | 53 | — | — | — | |
| MarginBudget size=b=1%, Backbone=ResNet1012026.05 | 55 | — | — | — | |
| RTSBudget size=b=5%, Backbone=ResNet1012026.05 | 55 | — | — | — | |
| RTSBudget size=b=3%, Backbone=ResNet1012026.05 | 56 | — | — | — | |
| CAFDBudget size=b=12%, Backbone=ResNet1012026.05 | 56 | — | — | — | |
| VanillaBudget size=b=7%, Backbone=ResNet1012026.05 | 57 | — | — | — | |
| DATISBudget size=b=7%, Backbone=ResNet1012026.05 | 58 | — | — | — | |
| DeepGiniBudget size=b=7%, Backbone=ResNet1012026.05 | 59 | — | — | — | |
| VanillaBudget size=b=5%, Backbone=ResNet1012026.05 | 60 | — | — | — | |
| CAFDBudget size=b=10%, Backbone=ResNet1012026.05 | 60 | — | — | — | |
| DATISBudget size=b=5%, Backbone=ResNet1012026.05 | 62 | — | — | — | |
| DeepGiniBudget size=b=5%, Backbone=ResNet1012026.05 | 63 | — | — | — | |
| VanillaBudget size=b=3%, Backbone=ResNet1012026.05 | 65 | — | — | — | |
| CAFDBudget size=b=7%, Backbone=ResNet1012026.05 | 65 | — | — | — | |
| RTSBudget size=b=1%, Backbone=ResNet1012026.05 | 66 | — | — | — | |
| DATISBudget size=b=3%, Backbone=ResNet1012026.05 | 66 | — | — | — | |
| DeepGiniBudget size=b=3%, Backbone=ResNet1012026.05 | 67 | — | — | — | |
| CAFDBudget size=b=5%, Backbone=ResNet1012026.05 | 69 | — | — | — | |
| VanillaBudget size=b=1%, Backbone=ResNet1012026.05 | 74 | — | — | — | |
| CAFDBudget size=b=3%, Backbone=ResNet1012026.05 | 74 | — | — | — | |
| DATISBudget size=b=1%, Backbone=ResNet1012026.05 | 75 | — | — | — | |
| DeepGiniBudget size=b=1%, Backbone=ResNet1012026.05 | 76 | — | — | — | |
| CAFDBudget size=b=1%, Backbone=ResNet1012026.05 | 81 | — | — | — | |
| BayesVLMBackbone=CLIP ViT-B/322025.11 | — | 79.41 | 86.05 | 72.83 | |
| BayesVLMBackbone=SigLIP ViT-B/162025.11 | — | 70.06 | 87.84 | 83.01 | |
| EntropyBackbone=CLIP ViT-B/322025.11 | — | 78.27 | 86.01 | 76.75 | |
| EntropyBackbone=CLIP ViT-B/162025.11 | — | 83.46 | 93.87 | 68.64 | |
| EntropyBackbone=SigLIP ViT-B/162025.11 | — | 83.46 | 93.87 | 68.64 | |
| MaxCosineBackbone=CLIP ViT-B/322025.11 | — | 65.79 | 74.58 | 86.4 | |
| MaxCosineBackbone=CLIP ViT-B/162025.11 | — | 70.2 | 87.04 | 82.31 | |
| MaxCosineBackbone=SigLIP ViT-B/162025.11 | — | 70.2 | 87.04 | 82.31 | |
| MaxSoftmaxBackbone=CLIP ViT-B/322025.11 | — | 80.71 | 87.43 | 71.38 | |
| MaxSoftmaxBackbone=CLIP ViT-B/162025.11 | — | 84.21 | 94.1 | 65.3 | |
| MaxSoftmaxBackbone=SigLIP ViT-B/162025.11 | — | 84.21 | 94.1 | 65.3 | |
| OursBackbone=CLIP ViT-B/322025.11 | — | 87.52 | 91.61 | 55.26 | |
| OursBackbone=CLIP ViT-B/162025.11 | — | 88.01 | 93.23 | 54.1 | |
| OursBackbone=SigLIP ViT-B/162025.11 | — | 88.57 | 95.85 | 48.87 | |
| Ours-DBackbone=CLIP ViT-B/32, Covariance=Diagonal2025.11 | — | 85.8 | 90.53 | 59.02 | |
| Ours-DBackbone=CLIP ViT-B/16, Covariance=Diagonal2025.11 | — | 86.8 | 92.57 | 56.57 | |
| Ours-DBackbone=SigLIP ViT-B/16, Covariance=Diagonal2025.11 | — | 86.17 | 94.8 | 56.88 | |
| PCME++Backbone=CLIP ViT-B/322025.11 | — | 77.31 | 90.68 | 71.82 | |
| ProbVLMBackbone=CLIP ViT-B/322025.11 | — | 50.18 | 61.79 | 95.31 | |
| ProLIPBackbone=CLIP ViT-B/162025.11 | — | 70.57 | 85.77 | 81.23 | |
| TempScalingBackbone=CLIP ViT-B/322025.11 | — | 80.63 | 87.37 | 71.64 | |
| TempScalingBackbone=CLIP ViT-B/162025.11 | — | 81.28 | 89.87 | 72.16 | |
| TempScalingBackbone=SigLIP ViT-B/162025.11 | — | 84.12 | 94.06 | 66.06 | |
| TracInBackbone=ResNet-18, Parameter-Weighted=false2025.06 | — | 54.4 | — | — | |
| TracInBackbone=ResNet-18, Parameter-Weighted=true2025.06 | — | 61.46 | — | — | |
| TracInBackbone=ViT-B/16, Parameter-Weighted=false2025.06 | — | 71.27 | — | — | |
| TracInBackbone=ViT-B/16, Parameter-Weighted=true2025.06 | — | 83.58 | — | — | |
| TRAKBackbone=ResNet-18, Parameter-Weighted=false2025.06 | — | 59.66 | — | — | |
| TRAKBackbone=ResNet-18, Parameter-Weighted=true2025.06 | — | 67.72 | — | — | |
| TRAKBackbone=ViT-B/16, Parameter-Weighted=false2025.06 | — | 80.08 | — | — | |
| TRAKBackbone=ViT-B/16, Parameter-Weighted=true2025.06 | — | 83.48 | — | — | |
| TrustVLM*Backbone=CLIP ViT-B/16, Extra features=DinoV22025.11 | — | 83.39 | 90.43 | 75.67 | |
| TrustVLM*Backbone=SigLIP ViT-B/16, Extra features=DinoV22025.11 | — | 81.66 | 90.85 | 81.65 | |
| ZeroBackbone=CLIP ViT-B/162025.11 | — | 78.87 | 90.84 | 73.95 | |
| ZeroBackbone=SigLIP ViT-B/162025.11 | — | 79.43 | 93.55 | 68.65 |