Error Detection on Flowers102
99.38AuROCOurs-D
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ours-DBackbone=SigLIP ViT-B/16, Covariance=Diagonal2025.11 | 99.38 | 99.88 | 1.4 | — | |
| OursBackbone=SigLIP ViT-B/162025.11 | 98.81 | 99.77 | 1.6 | — | |
| OursBackbone=CLIP ViT-B/162025.11 | 98.25 | 99.15 | 2.47 | — | |
| OursBackbone=CLIP ViT-B/322025.11 | 97.59 | 98.68 | 4.86 | — | |
| Ours-DBackbone=CLIP ViT-B/16, Covariance=Diagonal2025.11 | 97.28 | 98.79 | 10.62 | — | |
| Ours-DBackbone=CLIP ViT-B/32, Covariance=Diagonal2025.11 | 96.96 | 98.43 | 14.79 | — | |
| CLS-MBackbone=ViT-B/16, Evaluation Protocol=16-shot Prompt Learning2024.05 | 95.69 | — | — | — | |
| CLS-EBackbone=ViT-B/16, Evaluation Protocol=16-shot Prompt Learning2024.05 | 94.96 | — | — | — | |
| MaxLogitBackbone=ViT-B/16, Evaluation Protocol=16-shot Prompt Learning2024.05 | 94.4 | — | — | — | |
| MSPBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 94.23 | — | — | — | |
| GradNormBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 93.75 | — | — | — | |
| TrustVLM*Backbone=SigLIP ViT-B/16, Extra features=DinoV22025.11 | 93.65 | 99.93 | 29.82 | — | |
| TempScalingBackbone=SigLIP ViT-B/162025.11 | 93.64 | 98.66 | 38.38 | — | |
| RMDSBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 93.58 | — | — | — | |
| MaxSoftmaxBackbone=CLIP ViT-B/162025.11 | 93.49 | 98.64 | 40.18 | — | |
| MaxSoftmaxBackbone=SigLIP ViT-B/162025.11 | 93.49 | 98.64 | 40.18 | — | |
| EnergyBackbone=ViT-B/16, Evaluation Protocol=16-shot Prompt Learning2024.05 | 93.36 | — | — | — | |
| EntropyBackbone=CLIP ViT-B/162025.11 | 93.09 | 98.55 | 36.77 | — | |
| EntropyBackbone=SigLIP ViT-B/162025.11 | 93.09 | 98.55 | 36.77 | — | |
| NegPromptBackbone=ViT-B/16, Evaluation Protocol=16-shot Prompt Learning2024.05 | 93 | — | — | — | |
| ASHBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 92.62 | — | — | — | |
| ReactBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 92.09 | — | — | — | |
| ODINBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 91.84 | — | — | — | |
| LoCoOp (GL-MCM)Backbone=ViT-B/16, Evaluation Protocol=16-shot Prompt Learning2024.05 | 89.54 | — | — | — | |
| TrustVLM*Backbone=CLIP ViT-B/16, Extra features=DinoV22025.11 | 89.18 | 99.86 | 34.43 | — | |
| VIMBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 88.03 | — | — | — | |
| MDSBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 86.68 | — | — | — | |
| ZeroBackbone=SigLIP ViT-B/162025.11 | 86.3 | 96.84 | 53.09 | — | |
| MaxSoftmaxBackbone=CLIP ViT-B/322025.11 | 86.26 | 92.21 | 61.11 | — | |
| TempScalingBackbone=CLIP ViT-B/322025.11 | 86.15 | 92.11 | 61.93 | — | |
| BayesVLMBackbone=CLIP ViT-B/322025.11 | 85.65 | 91.88 | 91.37 | — | |
| TempScalingBackbone=CLIP ViT-B/162025.11 | 85.41 | 92.6 | 60.58 | — | |
| EntropyBackbone=CLIP ViT-B/322025.11 | 85.22 | 91.56 | 64.7 | — | |
| NegLabelBackbone=ViT-B/16, Evaluation Protocol=Zero-shot CLIP2024.05 | 83.99 | — | — | — | |
| ZeroBackbone=CLIP ViT-B/162025.11 | 83.73 | 92.3 | 62.98 | — | |
| CLIPNBackbone=ViT-B/16, Evaluation Protocol=Zero-shot CLIP2024.05 | 82.58 | — | — | — | |
| ProLIPBackbone=CLIP ViT-B/162025.11 | 80.67 | 92.82 | 59.95 | — | |
| PCME++Backbone=CLIP ViT-B/322025.11 | 80.05 | 93.89 | 70.11 | — | |
| MaxCosineBackbone=CLIP ViT-B/162025.11 | 79.98 | 94.32 | 64.23 | — | |
| MaxCosineBackbone=SigLIP ViT-B/162025.11 | 79.98 | 94.32 | 64.23 | — | |
| MCMBackbone=ViT-B/16, Evaluation Protocol=Zero-shot CLIP2024.05 | 78.57 | — | — | — | |
| AdaNegBackbone=ViT-B/16, Evaluation Protocol=Zero-shot CLIP2024.05 | 74.12 | — | — | — | |
| MaxCosineBackbone=CLIP ViT-B/322025.11 | 72.71 | 82.61 | 80.85 | — | |
| BayesVLMBackbone=SigLIP ViT-B/162025.11 | 70.59 | 90.49 | 85.19 | — | |
| KNNBackbone=ViT-B/16, Evaluation Protocol=Fully Fine-tuned2024.05 | 52.5 | — | — | — | |
| ProbVLMBackbone=CLIP ViT-B/322025.11 | 43.05 | 53.74 | 97.2 | — | |
| MSP (Cls-Preserving)Model=ViT-L2026.04 | — | — | — | 0 | |
| MSP (Cls-Preserving)Model=ViT-B2026.04 | — | — | — | 0 | |
| MSP (Disc-Optimized)Model=ViT-L2026.04 | — | — | — | 19.7 | |
| MSP (Disc-Optimized)Model=ViT-B2026.04 | — | — | — | 1.5 | |
| VIM (Cls-Preserving)Model=ViT-L2026.04 | — | — | — | 0 | |
| VIM (Cls-Preserving)Model=ViT-B2026.04 | — | — | — | 0 | |
| VIM (Disc-Optimized)Model=ViT-L2026.04 | — | — | — | 0.2 | |
| VIM (Disc-Optimized)Model=ViT-B2026.04 | — | — | — | 4.5 |