🔎 HCLR-Net

Introduction

📝 Note

• This is a personal implementation of HCLR-Net, based on the official implementation.
• Compared with the official implementation, this implementation supports full-size evaluation instead of images with fixed resolution.

1. Installation

• Clone the repo:

  git clone https://github.com/xigua7105/HCLR-Net.git
  cd HCLR-Net
  

• Environment setup: HCLR-Net works with Python 3.8+ and PyTorch 2.0+.

  conda create -n HCLR python=3.8
  conda activate HCLR
  pip install -r requirements.txt
  

2. QuickStart

• To quickly enhance an underwater image, run:

   python demo_image.py
  

• To quickly enhance an underwater video, run:

   python demo_video.py
  

• Train your model:

  python run.py --c configs/train/HCLR-Net.yaml --mode train
  

• Test your model:

  python run.py --c configs/test/HCLR-Net-Test.yaml --mode test
  

Implementation Details

💿 Datasets

💡 We expect the directory structure to be the following:

• path/to/dataset/
  |---train
  |   |---LSUI/
  |   |   |---input
  |   |   |---target
  |   |---UIEB/
  |       |---input
  |       |---target
  |
  |---test/
  |   |---LSUI/
  |   |   |---input
  |   |   |---target
  |   |---UIEB/
  |       |---input
  |       |---target
  |   |---EUVP/
  |       |---input
  |       |---target
  

• ⬇️ Download dataset from LSUI, UIEB and EUVP.

📝 Configs

• More details see this and DLTrainer.

👍 Acknowledgements

• This repo is mainly built on open repo DLTrainer and HCLR-Net official implementation.
• Thanks to these open-source repositories for their contribution to advancing the development of underwater image enhancement.