Exposure Correction

Restore over-exposure images based on Zero-DCE

Related Works  •  Contributors  •  Pipeline  •  Results  •  Environment  •  Installation

Note

Exposure Correction is used to adjust the brightness of photos or images to achieve an ideal exposure level, correcting detail loss caused by overexposure (too bright) or underexposure (too dark), thereby enhancing image quality and visual effects.
Our work combines the methods from three papers, modifying the network framework to achieve effective exposure correction.

Related Works

• Zero DCE, CVPR 2020
• Learning Multi-Scale Photo Exposure Correction, CVPR 2021
• Reversed and Fused Zero-DCE

Contributors

Project Explanation Video

Video

Pipeline

Our pipeline consists of three main components:

1. Multi-scale Processing: Input image is decomposed into n-level Laplacian pyramid
2. Progressive Enhancement:
   • Each level is processed by a dedicated DCE-Net
   • Forward enhancement: Apply curve parameters A
   • Backward reasoning: Apply inverse parameters -A
   • Add Laplacian details to enhanced result
3. Cascading Structure: Output from each level serves as input to the next level

Results

Over-exposure

Under-exposure

Environment

Caution

We are using the following environment to develop this project.

• NVIDIA GeForce RTX 4090
• Ubuntu 24.04 LTS
• python = 3.7
• pytorch = 1.13.1
• torchvision = 0.14.1
• torchaudio = 0.13.1
• Cuda = 11.7
• OpenCV

Installation

1. Clone the repository:

   git clone https://github.com/ChuEating1005/Exposure-Correction
   cd Exposure-Correction

2. Install dependencies:

   pip install -r requirements.txt

Project Structure

Zero-DCE/
├── data/               # Dataset directory
├── models/            # Model architectures
├── train/             # Training scripts
├── test/              # Testing scripts
├── utils/             # Utility functions
└── snapshots/         # Model weights

License

This project is licensed under the MIT License - see the LICENSE file for details.