LoopExpose

Try our model on

Introduction

This repository is the official implementation of "LoopExpose: An Unsupervised Framework for Arbitrary-Length Exposure Correction"

Ao Li¹, Chen Chen⁴, Zhenyu Wang^2*, Tao Huang², Fangfang Wu³, Weisheng Dong ¹

¹School of Artificial Intelligence, Xidian University

²Hangzhou Institute of technology, Xidian University

³School of Computer Science and Technology, Xidian University

⁴Dalian University of Technology

*: Corresponding Author.

Datasets

• MSEC https://github.com/mahmoudnafifi/Exposure_Correction

• UEC https://github.com/BeyondHeaven/uec_code

  After downloading the above two datasets, you can run the files in the data_utils folder to reorganize the datasets.

Environment

OS: Ubuntu 20.04.6

python == 3.9.19

torch == 2.4.1

opencv == 4.10.0

This model is trained on an RTX 4090 GPU, taking about a day and occupies approximately 24GB of memory.

Usage

train

Please refer to Main.py for options information.

python Main.py

test

Checkpoints are released at ckpts.

python Test.py

If you have any questions about the code, please email me directly : liaoxdu@foxmail.com or ali_0607@stu.xidian.edu.cn .

Acknowledgment and Future works

This implementation is based on CoTF、LACT、MEFNet and OpenCV. In the future, we will incorporate more exposure correction models and exposure fusion models into our framework. Everyone is also welcome to contribute to this project.