Skip to content

Latest commit

 

History

History
73 lines (54 loc) · 3.31 KB

README.md

File metadata and controls

73 lines (54 loc) · 3.31 KB
Raw

AllWeatherNet (ICPR 2024 Best Student Paper)

Unified Image Enhancement for Autonomous Driving under Adverse Weather and Lowlight Conditions

AllWeatherNet is an advanced image enhancement framework designed to improve the visibility of images captured in challenging weather conditions, such as snow, rain, fog, and low-light environments. This solution is particularly tailored for autonomous driving applications, ensuring safety and performance in diverse environments.

Original Input Original Input 2

Key Features

  • Unified Enhancement: Enhances images captured under various adverse weather conditions, including snowy, rainy, foggy, and nighttime scenarios.
  • Scaled-Illumination Attention: Employs a robust scaled-illumination attention mechanism to maintain focus on the road across different conditions.
  • Hierarchical Discrimination: Utilizes hierarchical patch-level discrimination at scene, object, and texture levels for more effective enhancement.

Architecture

Environment Setup

To set up the required environment, run:

pip install -r requirements.txt

Dataset Preparation

  1. Download the dataset from the ACDC official website.
  2. Organize the dataset in the following structure: 
    ├── ACDC
│   ├── trainA  # Contains adverse weather images
│   └── trainB  # Contains normal weather images

Demo Instructions

  1. Download the Pretrained Model:
    Download the pretrained model from this link and place it in the checkpoints folder within the repository.

  2. Set the Demo Image Path:
    Specify the path to your demo image by setting the dataroot variable in script.py. The image can either be the original or a downsampled version from the original dataset.

  3. Run the Script:
    Execute the script using the following command:

    python scripts/script.py --predict

Acknowledgements

This project is inspired by EnlightenGAN. We greatly appreciate the authors for their outstanding contributions.

Citation

If you find this work useful, please cite:

@inbook{Qian_2024,
   title={AllWeather-Net: Unified Image Enhancement for Autonomous Driving Under Adverse Weather and Low-Light Conditions},
   ISBN={9783031781131},
   ISSN={1611-3349},
   url={http://dx.doi.org/10.1007/978-3-031-78113-1_11},
   DOI={10.1007/978-3-031-78113-1_11},
   booktitle={Pattern Recognition},
   publisher={Springer Nature Switzerland},
   author={Qian, Chenghao and Rezaei, Mahdi and Anwar, Saeed and Li, Wenjing and Hussain, Tanveer and Azarmi, Mohsen and Wang, Wei},
   year={2024},
   month=dec, pages={151–166} }

To-Do List

  • Release test code
  • Clean and refine training code
  • Add more documentation and tutorials

Views