This repo contains the official implementation of the IJCAI 2022 paper 'Domain Adversarial Learning for Color Constancy', the code will be available after the IJCAI 2022 meeting.

Domain Adversarial Learning for Color Constancy

Zhifeng Zhang, Xuejing Kang, Anlong Ming

School of Computer Science, Beijing University of Posts and Telecommunications

1: Required Environment Installation

Please use the command pip install -r requirements.txt or manually install the following packages.

• torch==1.9.0+cu111
• numpy==1.20.1
• matplotlib==3.3.4
• opencv_python==4.5.3.56
• torchsummary==1.5.1
• scipy==1.6.2
• torchvision==0.10.0+cu111

2: Data Preprocessing

For details on RAW image preprocessing principle, please refer to RAW_preprocessing_MATLAB and RAW_preprocessing_Python.

Color Checker Data preprocessing:

• Download the dataset on: https://www2.cs.sfu.ca/~colour/data/shi_gehler/
• Store the download file in: /dataset/colorconstancy/colorchecker2010/
• Set the path of the output file, such as: /home/*/data/CC_full_size/ or /home/*/data/CC_resize/
• Run the ./dataset/color_constancy_data_process_all.py code

Cube+ Data preprocessing

• Download the dataset on: https://ipg.fer.hr/ipg/resources/color_constancy
• Store the download file in: /dataset/colorconstancy/Cube/
• Set the path of the output file, such as: /home/*/data/Cube_full_size/ or /home/*/data/Cube_resize/
• Run the ./dataset/color_constancy_data_process_all.py code

NUS Data preprocessing

• Download the dataset on: https://cvil.eecs.yorku.ca/projects/public_html/illuminant/illuminant.html
• Store the download file in: /dataset/colorconstancy/NUS/
• Set the path of the output file, such as: /home/*/data/NUS_full_size/ or /home/*/data/NUS_resize/
• Run the ./dataset/color_constancy_data_process_all.py code

Summary

• If you have done the above, run the ./dataset/data_read.py for a simple test.
• Run ./dataset/data_loder.py to check that the data loader is running correctly.
• Refer to the ./dataset/fold_dalcc.py to check the three cross-validation. The fold for cross-validation can be found in ./data_fold/

3: Network Model Checking

• Download the Alexnet model parameter that pre-trained on ImageNet, and run the ./model/alexnet.py to check the AlexNet model.
• Run the ./model/DALCC.py to check the DALCC model.

4: Train and Test

• Run the ./train.py to train the DALCC model.
• Run the ./test.py to test the DALCC model.

5: Pretrained model

• You can find the pretrained model on Cube+ and NUS-8 on Pretrained Model.

6: Visualization of our DALCC.