This repository contains the source code and supplementary materials for the paper titled ALEN: An Adaptive Dual-Approach for Enhancing Uniform and Non-Uniform Low-Light Images. This research focuses on low-light images enhancement.
- opencv-python == 4.9.0.80
- scikit-image == 0.22.0
- numpy == 1.24.3
- torch == 2.3.0+cu118
- Pillow == 10.2.0
- tqdm == 4.65.0
- natsort == 8.4.0
- torchvision == 0.18.0+cu118
To test the model, follow these steps:
-
Download the code and pretrained weights from this link.
-
Place your images to be enhanced in the ./1_Input directory.
-
Run the code with the following command:
python inference.py
-
The enhanced images will be saved in the ./2_Output directory.
This section describes the datasets used to train and evaluate the performance of ALEN: Adaptive Light Enhancement Network for low-light image enhancement.
The following public datasets were used to train the ALEN model. These datasets contain images with global and local illumination variations, necessary for effective classification and enhancement:
| Dataset | Description | Number of Images | Type | Resources |
|---|---|---|---|---|
| GLI | Global-Local Illumination | 2,000 | Paired Classification | Dataset |
| HDR+ | High Dynamic Range Plus | 922 | Paired Enhancement | Paper/Dataset |
| SLL | Synthetic Low-Light | 22,472 | Paired Enhancement | Paper/Dataset |
| MIT | MIT-Adobe FiveK | 5,000 | Paired Enhancement | Paper/Dataset |
To evaluate the overall performance and generalization ability of ALEN, we used various datasets representing real-world scenarios:
| Dataset | Description | Number of Images | Type | Resources |
|---|---|---|---|---|
| DIS | Diverse Illumination Scene | 10 | Unpaired Enhancement | Dataset |
| LSRW | Large-Scale Real-World | 735 | Paired Enhancement | Paper/Dataset |
| UHD-LOL4k | Ultra-High Definition LOw-Light 4K | 735 | Paired Enhancement | Paper/Dataset |
| DICM | -------------------- | 69 | Unpaired Enhancement | Paper/Dataset |
| LIME | Low-light Image Enhancement | 10 | Unpaired Enhancement | Paper/Dataset |
| MEF | Multi-Exposure Fusion | 17 | Unpaired Enhancement | Paper/Dataset |
| NPE | Naturalness Preserved Enhancement | 8 | Unpaired Enhancement | Paper/Dataset |
| TM-DIED | The Most Difficult Image Enhancement Dataset | 222 | Unpaired Enhancement | Dataset |