LLIE

Implementation of 2 different algorithms:

• Zero-DCE for low-light image enhancement [link]
• Shallow camera pipeline for night photography rendering (IVL) [link]

The first one is an efficient unsupervised deep learning method, while the last one is an algorithm based on traditional image processing techniques.

These algorithm were tested on the LOL dataset, which you need to download (under the Desktop dir) in order to train Zero-DCE and test them.

Requirements

The code was tested with Python 3.10.12.

Install requirements:

pip install -r requirements.txt

Test IVL

In order to test this algorithm:

N.B. Change the path of the image to be tested in methods/ivl.py file.

cd methods/
python ivl.py

Train Zero-DCE

The training will save checkpoints under a ckpt/ folder under DCE directory.

N.B. Change the path to the dataset in methods/DCE/main.py file.

cd methods/DCE/
python main.py --train True

Compare methods

Then, we can compare the 2 methods described above, and a naive gamma correction. The script will save the images enhanced under a results/ folder, and output a comparision about: latencies, and several FIQA metrics (PSNR, MSE, MAE and SSIM).

N.B. You need to train the Zero-DCE model before.

python main.py --image 2.png