Bobby Newmark · 2022-05-09 15:03
Attached image is the source and an upscale output by ESRGAN (using the "universal upscaler (detailed)" model) via cupscale. Pretty amazing. Aside from some minor details, it looks almost native.
https://github.com/n00mkrad/cupscale
https://upscale.wiki/wiki/Model_Database
FILLER · 2022-05-09 17:48
Cool, too bad 99% of this soykaf is CUDA only. Here are some Vapoursynth plugins for non-plebeians:
https://github.com/rlaphoenix/VSGAN
https://github.com/HolyWu/vs-waifu2x-ncnn-vulkan
https://github.com/HolyWu/vs-realesrgan
https://github.com/HolyWu/vs-rife
https://github.com/HolyWu/vs-basicvsrpp
https://github.com/HolyWu/vs-dpir
Undisclosed · 2022-05-10 04:17
>>31136
I've been meaning to write something like yatta IVTC but for vapoursynth or avisynth+. It'd be very useful to have this running as part of an avs script, lot of cases where there were segments with a lot of mosquito noise I had no good way to clean up back in the day. Some models might easily outdo eg NNEDI3 in quality and speed.