Supplementary File for Finding Discriminative Filters for Specific Degradations in Blind Super-Resolution

Neural Information Processing Systems 

More experiments and analyses of masking discovered filters in SRCNN-style and SRResNet networks are provided in Section 2 and Section 3, respectively. The impact of choosing different baseline models in FAIG is discussed in Section 4. We provide more results of degradation prediction in Section 5. Finally, we present a controllable adjustment of restoration strength with the discovered degradation-specific filters in Section 6. Furthermore, we include the core codes of FAIG in this supplementary file.