Juan C. Benito, Daniel Feijoo, Alvaro Garcia, Marcos V. Conde (CIDAUT AI and University of Würzburg)

Abstract: Low-Light Image Enhancement (LLIE) is a key task in computational photography and imaging. The problem of enhancing images captured during night or in dark environments has been well-studied in the image signal processing literature. However, current deep learning-based solutions struggle with efficiency and robustness in real-world scenarios (e.g. scenes with noise, saturated pixels, bad illumination). We propose a lightweight neural network that combines image processing in the frequency and spatial domains. Our method, FLOL+, is one of the fastest models for this task, achieving state-of-the-art results on popular real scenes datasets such as LOL and LSRW. Moreover, we are able to process 1080p images under 12ms. Our code and models will be open-source.

Input	UHDFour	FLOL (ours)
Input	UHDFour	FLOL (ours)

• Python == 3.10.12
• PyTorch == 2.1.0
• CUDA == 12.1
• Other required packages in requirements.txt

# Clone this repository
git clone https://github.com/cidautai/FLOL.git
cd FLOL

# Create python environment and activate it
python3 -m venv venv_FLOL
source venv_FLOL/bin/activate

# Install python dependencies
pip install -r requirements.txt

The datasets used for training and/or evaluation are:

You can download LOLv2-Real and UHD-LL datasets and put them on the /datasets folder for testing.

We present results in different datasets for FLOL+.

To check our results you could run the evaluation of DarkIR in each of the datasets:

• Run python evaluation.py --config ./options/LOLv2-Real.yml on your terminal to obtain PSNR and SSIM metrics. Default is UHD-LL.

• Run python lpips_metric.py -g /LSRW_GroundTruthImages_path -p /LSRW_predictedimages -e .jpg on your terminal to obtain LPIPS value. (LSRW predicted images are obtained by using LOLv2-Real weight file)

You can process the entire set of test images of provided datasets by running:

• Run python inference.py --config ./options/LOLv2-Real.yml (UHD-LL is set by default)

Processed images will be saved in ./results/dataset_selected/.

LSRW-Huawei

Input	FECNet	SNR-Net	FourLLIE	FLOL (ours)	Ground Truth

LSRW-Nikon

Input	MIRNet	RUAS	EnGAN	FLOL (ours)	Ground Truth

UHD-LL

Input	UHDFour	FLOL (ours)	Ground Truth

This work is licensed under the MIT License.

If you have any questions, please contact juaben@cidaut.es and marcos.conde@uni-wuerzburg.de