About generalizing to unseen degradation type?

[yc-cui]

Thanks for your impressive work! In the appendix, the types of degradation observed in Table 5 and Figure 17 closely resemble those of the trained degradation. Have you considered applying the DA-CLIP model to a broader range of tasks to test the generalization?

[Algolzw]

Currently, we only collected 10 datasets for training and testing and it is hard to apply this model for other degradations. We also tried to perform general restoration in the wild (with small and mild degradations) in this paper but I have to say using pretrained stable-diffusion leads to a better generalization. The large-scale dataset may be the most important thing for real-world applications!