Error Detection on ImageNet V2 (test)
86.19AuROCOurs
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OursBackbone=CLIP ViT-162025.11 | 86.19 | 92.42 | 59 | |
| Ours-DBackbone=CLIP ViT-162025.11 | 85.32 | 91.91 | 62.27 | |
| TrustVLMBackbone=CLIP ViT-162025.11 | 83.03 | 89.68 | 78.76 | |
| TempScalingBackbone=CLIP ViT-162025.11 | 81.53 | 89.95 | 70.43 | |
| MaxSoftmaxBackbone=CLIP ViT-162025.11 | 81.3 | 89.83 | 71 | |
| EntropyBackbone=CLIP ViT-162025.11 | 78.78 | 88.62 | 78.08 | |
| MaxCosineBackbone=CLIP ViT-162025.11 | 67.84 | 79.59 | 83.55 |