RRDNet
RRDNet copied to clipboard
aaaaangel
Implementation of "Zero-Shot Restoration of Underexposed Images via Robust Retinex Decomposition, International Conference on Multimedia and Expo, 2020"
I am interested in the article, and I think it's a good job. I want to know whether you retain the NIQE model, my test data are bigger than article?
Hello, I have two questions about your loss functions. In your paper, the Illumination-guided noise estimation loss is shown as:  But in your codes, wn(x) represents illumination and wr(x)...
> def noise_loss(image, illumination, reflectance, noise): weight_illu = illumination weight_illu.detach() loss = weight_illu*noise return torch.norm(loss, 2)
Although the enhanced results is somehow underexposed, it's good enough. However, more work may need to be done upon speed.
