Denoising before upscaling: why the order matters for compressed video
PyTorch Forums
Denoising before upscaling: why the order matters for compressed video
I kept running into the same failure while building a browser-based upscaler: models that enlarge first also enlarge the compression blocks. The order that worked for me: 1. Denoise (target compression artifacts, not sensor grain) 2. Enlarge at most 2x per axis 3. Inspect at 100% crop 4. Grade last Past 2x the model starts inventing detail — faces go waxy, small text becomes strokes that look like letters. Compression noise is structural (a regular block grid), not random, so treating it a...
0 comments
No comments yet.