Error detection on ImageNet-A (test)
0.7614AuROCMaxSoftmax
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MaxSoftmaxBackbone=CLIP ViT-162025.11 | 0.7614 | 0.7463 | 0.7774 | |
| TempScalingBackbone=CLIP ViT-162025.11 | 0.7612 | 0.7484 | 0.782 | |
| EntropyBackbone=CLIP ViT-162025.11 | 0.7482 | 0.734 | 0.7766 | |
| Ours-DBackbone=CLIP ViT-162025.11 | 0.7034 | 0.6753 | 0.7799 | |
| OursBackbone=CLIP ViT-162025.11 | 0.6883 | 0.6474 | 0.7838 | |
| MaxCosineBackbone=CLIP ViT-162025.11 | 0.6429 | 0.6095 | 0.8667 | |
| TrustVLMBackbone=CLIP ViT-162025.11 | 0.6428 | 0.4953 | 0.9292 |