This repo is an unofficial implement of paper "Noise-As-Clean: Learning Self-supervised Denoising from the Corrupted Image". The original github repo is here but very complicated as I think.
The denoising result I get is much lower than that is reported in the paper. That is probably resulted by:
- I only train the model on single image.
- I donot use much data argumentation.
If you can improve the code, please create pull request, or contact me via xinge.yang@kaust.edu.sa
python nac_single_img.py
NAC provides a novel strategy for single image blind denoising, but I think it is a biased estimation and the proof in the paper is not precise.