Awesome

EEMEFN: Low-Light Image Enhancement via Edge-Enhanced Multi-Exposure Fusion Network

Introduction

This project page provides TensorFlow 1.X code that implements the following AAAI2019 paper:

Title: "EEMEFN: Low-Light Image Enhancement via Edge-Enhanced Multi-Exposure Fusion Network"

Paper: https://aaai.org/ojs/index.php/AAAI/article/view/7013/6867

How to use

conda env create --prefix ./env --file environment.yml
conda activate .env
pip install h5py
pip install rawpy

Download pretrained model

Sony and Fuji model

# model directory structure
result
|-- Sony_edge
|-- Sony_merge
|-- Fuji_edge
└-- Fuji_merge

Process data

Please download raw dataset from https://github.com/cchen156/Learning-to-See-in-the-Dark
And delete misalignment images
Then process data as follows (about two hours)

python dataset.py

# dataset directory structure
dataset
|-- Fuji
|   |-- fuji_long.hdf5
|   |-- fuji_short.hdf5
|   |-- long
|   └-- short
└-- Sony
    |-- sony_long.hdf5
    |-- sony_short.hdf5
    |-- long
    └-- short

Training

Please modify the name of target dataset first

python train_fusion.py
python train_merge.py

Evaluation

python test.py

Performance

License

This code is released under the MIT License (refer to the LICENSE file for details).