Deep neural network project to enlighten images based on deep curve estimation. This is an unoficial implementation of this paper: https://openaccess.thecvf.com/content_CVPR_2020/papers/Guo_Zero-Reference_Deep_Curve_Estimation_for_Low-Light_Image_Enhancement_CVPR_2020_paper.pdf.
Make sure to download the training and testing data from https://drive.google.com/file/d/1GAB3uGsmAyLgtDBDONbil08vVu5wJcG3/view. (SICE Dataset) and unrar it into following folder called SICE, and have a folder called result and original.
Otherwise just make a copy of these folders in your gdrive:
SICE dataset: https://drive.google.com/drive/folders/1AFKwrOztjCzgZKPzcZ_2pCNoRn8GvOvX?usp=sharing
then make a empty folder called Result and Original.
The result will have the final enhanced image and orginal will have the the unhanced original.
The results are very promising and it is definitely better than just enlightning each pixel by a factor of 2. This Deepnetwork enhances the bright parts of an image by a less amount and the dark part by a large amount.
Here are the links to my results, when I tried with another different dataset(You are free to experiment & play with any dataset you like):
Original: https://drive.google.com/drive/folders/1gDwNqXtRdJ1kej3sBPPXoNNFsIEVOQeS?usp=sharing
Result: https://drive.google.com/drive/folders/1gDwNqXtRdJ1kej3sBPPXoNNFsIEVOQeS?usp=sharing