Status on neural/model training for upscaling?

There have not been any significant updates on neural/model training for upscaling for quite some time. A new face recognition model has been coveted as well.

Upscaling is the main thing I use TVAI for, and I’m concerned about the fact the the application has gotten a main version upgrade and many releases now which are seem to be trying to get it to a usable, stable state, while model training seems to have been put on hold.

Some people I know who have been using TVAI have started switching to xSRGAN frameworks that are available since they seem to be improving to the point where they surpass TVAI which hasn’t had any model improvements for upscaling since version 2.x.

Can we please get a statement or clarification on where model training for upscaling currently stands?


What program you referring to they switching to?

According to my experience, SRGAN cannot be directly used for video upscaling, it will destroy the details of some images and change the viewing experience of some pictures.

Unfortunately I don’t know the exact details but I see results from people that surpass Topaz, and they’re saying they are using SRGAN, but don’t want to provide further details (which framework, etc.). The main point is, I’d like to hear from Topaz as to where neural/model training stands.

The original SRGAN will remove the input grain, blur and some lines, etc. This kind of excessive sharpening will not be welcomed by everyone, and the enhancement of the video should also consider the respect of the source content.