- Use an end-to-end learned based method to do the image enhancement task: image quality enhancement, extreme low-light image enhancement task
- Meanwhile, solve the problem of the artifacts in the images produced by the state-of-the-art methods (DPED and SID).
- Use U-Net GAN as the backbone
- To further remove artifacts, our proposed model include two components to extract the global information(histogram, brightness, scene categories, etc.) in the images
- Self attention modules
- Global feature vector
- In the task of image quality enhancement:
- In the task of extreme low-light image enhancement:
https://docs.google.com/presentation/d/1lXiYRm-Tf6IlyN0lCGF2MXJr4gIzHSIpvdx-YEA392c/edit?usp=sharing
(https://drive.google.com/open?id=1Bj9PABfD5eftHeM4ifE4s8UAmvpVn6Qj):
- move model/dped/* to ./DPED/models/
- move model/sid/* to ./sid/sid_w_sa/