Upscaling 3D video?

Let’s say I had some fairly low-resolution 3D footage, is there a way I could prepare it to upscale to a higher resolution using VEAI? And if not; would it be possible to have a model trained for upscaling low-resolution 3D footage? I think this could be very useful for those that need it, especially since not much 3D footage is produced anymore in the grand scheme of things, and old 3D footage tends to be a fairly low resolution so it would be great if there was a way to do this, and ideally do it fairly easily.

1 Like

Is it anaglyph footage (the kind that requires red/blue glasses)? If so, maybe you could use a color filter to separate the two images, upscale each separately, then re-join them?

This would be interesting to know…
Stereoscopic footage from K1pro or EVO for example

1 Like

I am so sorry, I completely forgot to mention the method used. It’s side by side, but my worry is that the ai might upscale each side slightly differently causing mismatch

Have you tried processing the 3D sbs footage on VEAI and seeing if the results are to your liking? A 3D sbs video file is just a regular video file that has both the left eye footage and the right eye footage on each half of the same frame. VEAI would just process it same as a non-3D video file. I tried processing a 3D sbs video with Artemis Dehalo same scale denoise/deblock and it looked like it came out OK with the 3D effect still there.

Yeah I have tried it, on various models too, the 3D effect is there, but some of the depth information seems distorted, hence why I asked about methods used or possibly an ai model trained with 3D footage

This is a really good question. I’m not sure how Topaz uses AI to do it’s processing, but if it’s analyzing images and trying to recognize an object to reconstruct it better, then side by side or any other method that has some sort of image manipulation already would not produce accurate results. Unless you input a pixel aspect for the side by side, it’s not a workable image to begin with. Better to separate the eyes and process each individually. Chances are that the reconstruction would be pretty similar, and less likely to artifact that creating rivalries.

1 Like

I’ve use VEAI to upscale Top/Bottom formatted stereoscopic 360º video from 4K to 8K and it works perfectly.