For advanced CV project in University of Houston. Goals:
- reimplement the state-of-art paper "Deep Retinex Decomposition for Low-Light Enhancement".
- test it on acquired low light faces.
- run a pretrained face&landmark detection model to test improvement on low light faces.
Here is the paper website: https://daooshee.github.io/BMVC2018website/
The author also has published their code and dataset. You can find it on above website.
Final presentation
Final report
lowlight.ipynb: implementation of Retinex-Net
Traditional_enhance.ipynb: implemenation of Gamma enahncement
landmark.ipynb: face and landmark deteciton
This work is done by tensorflow 1.10 + Python3.6.6
Training can be finished in 1 hour, with a GTX1060(6G RAM).
{low Light, normal light, reflectance, illumination}
