sr-implicit-downsampled-fast

[Jinchen2028]

Thank you very much for sharing this excellent work, in which we noticed the use of "sr-implicit-downsampled-fast" in this work, and I would like to know what changes were made to make this "quick method" faster compared to "sr-implicit-downsampled" in LIIF?

[2y7c3]

We follow a prior implementation and utilize 1×1 convolutions to replace all the linear layers in SRNO, since they have a GPU-friendly data structure.