4 ms·

Would it not be possible to do the image processing on the original image to create several pixel masks which then draw in the high resolution image data? For

Would it not be possible to do the image processing on the original image to create several pixel masks which then draw in the high resolution image data? For instance, you take the source image at resolution n*n and do all processing at n/2 or n/4. Then at the moment you are going to composite the finished image you instead draw "pixels" from a source image which contains X,Y not RGB. Then you upscale the output image x2 (or x4) and replace each X,Y index with the relevant pixel from the source image (adding X%2,Y%2 or X%4 to the indexes to return to source resolution).

sniklaus 7y ago
That is called guided upsampling and we are in fact doing just that to get a better estimate of the scene geometry. Specifically, we estimate the scene geometry at a resolution of 256x256 pixels (for a 1:1 aspect ratio, it varies for other resolutions) and then use the original image as a guide to upsample this estimate to 1024x1024 pixels. So yes, it could be used to a greater extent but it needs more work to be able to do this.