Project Report
Abstract
Given a low-light image, the aim is to get a high quality, well-lit version of the image, while avoiding overexposure of parts of the scene. Our approach uses a combination of the frameworks used in the papers “Learning to See in the Dark” and “Burst photography for high dynamic range and low-light imaging on mobile cameras”. We use the U-Net neural network (used in [1]) to get a well-lit image from an input low-light image and apply tone mapping (as in [2]) to get the final image, which is not overexposed.
Dependencies
python 2.7 Tensorflow (>=1.1) Scipy Numpy Rawpy
Running the code
Run the command: python tune_Sony.py
References
[1] Learning to See in the Dark [2] Burst photography for high dynamic range and low-light imaging on mobile cameras