VEAI: Resize image before AI processing

There’s so many instances where an SD video has already been upscaled to HD, requiring us to shrink it back to the original resolution prior to processing with VEAI. I’ve ended up spending a lot of time resizing input videos just so the AI has the best chance of working. Sometimes I’ll make versions with Nearest Neighbor, Bilinear, and Bicubic resampling. The end result can be fantastic, but it does require that extra bit of planning (not to mention disk space). My feature request is for there to be a way to resize the input video within VEAI itself and to select the resampling method.

EDIT: Wait, why did this show up in Topaz Products instead of Feature Requests?

EDIT: Can somebody move this thread? I have no idea why it’s posted here. I was in Feature Requests when I wrote it.

Where do you want it moved to?

It would be useful if deep learning could be applied to essentially drop interpolated frequencies from the input; in other words the frequency domain equivalent of downscaling the image back to its original resolution prior to AI upscaling.

You can see a hand-crafted algorithm here that attempts to find reflections in the frequency domain to approximate the original, non-upscaled resolution of an image that has been interpolated: https://github.com/0x09/resdet