why

[zhaozw-szu]

[Knightzjz]

Hi, thanks for your attention, would you mind to be more specific rather than the "why"?
If you are referring to the error prediction of the NCL model on your images, you can check the similar answered question here. We currently do not load the fully trained NCL model, and the generalization might be then jeopardized.
If you are looking for a more powerful prototype, i.e. better generalization, you can reach and play our latest model through IML-ViT.