Near Out-of-Distribution Detection on CIFAR-10 (in) vs CIFAR-100 (out)
98.52Mahalanobis AUROCR50+ViT-B_16
Evaluation results
| Method | Links | ||
|---|---|---|---|
| R50+ViT-B_16Backbone=R50+ViT-B_16, Pre-trained=ImageNet-21k, In-distribution=CIFAR-10, Out-distribution=CIFAR-1002021.06 | 98.52 | 97.75 | |
| ViT-B_16Backbone=ViT-B_16, Pre-trained=ImageNet-21k, In-distribution=CIFAR-10, Out-distribution=CIFAR-1002021.06 | 98.42 | 97.68 | |
| MLP-Mixer-B_16Backbone=MLP-Mixer-B_16, Pre-trained=ImageNet-21k, In-distribution=CIFAR-10, Out-distribution=CIFAR-1002021.06 | 97.85 | 96.28 | |
| BiT-M R50x1Backbone=BiT-M R50x1, Pre-trained=ImageNet-21k, In-distribution=CIFAR-10, Out-distribution=CIFAR-1002021.06 | 95.52 | 85.87 | |
| BiT-M R101x3Backbone=BiT-M R101x3, Pre-trained=ImageNet-21k, In-distribution=CIFAR-10, Out-distribution=CIFAR-1002021.06 | 94.55 | 85.34 |