two-camera-white-balance

Code and data for the paper:

Leveraging the Availability of Two Cameras for Illuminant Estimation, CVPR 2021

Abdelrahman Abdelhamed, Abhijith Punnappurath, Michael S. Brown

Samsung Artificial Intelligence Center, Toronto, Canada

Goal

State-of-the-art illuminant estimation performance using two cameras.

More details in this Samsung Research blog post.

Presentation video at CVPR 2021

Maintainer(s)

• Abdelrahman Abdelhamed (a.abdelhamed@samsung.com; abdoukamel@gmail.com)

Committer(s)

• Abdelrahman Abdelhamed (a.abdelhamed@samsung.com; abdoukamel@gmail.com)

Datasets

• S20-Two-Camera-Dataset

• S20-Two-Camera-Dataset (metadata): Metadata_Image_Pairs.zip

• S20-Two-Camera-Dataset (augmented metadata): Metadata_Image_Pairs_Augment_99.zip

Tested environment

Ubuntu 18.04, Python 3.7, CUDA 11.2, cuDNN 8.1, TensorFlow 2.5

The code may work in other environments.

Setup

Install required packages and setup a virtual environment:

. ./scritps/setup.sh

Training and Testing

S20-Two-Camera-Dataset, augmented, 200 parameters

Training: Download and unzip Metadata_Image_Pairs_Augment_99.zip into ./data directory then run three-fold cross validation:

python -m jobs.train_s20_aug_200_3fold

Testing: Unzip Metadata_Image_Pairs.zip into the ./data directory then run the following command and feed in a comma-separated string of paths to models to be tested.

python -m jobs.test_s20_aug_200_3fold --test_model_paths <test_model_path_1,test_model_path_2,test_model_path_3>

Citation

If you use this code or the associated data, please cite the paper:

@InProceedings{Abdelhamed_2021_CVPR,
author = {Abdelhamed, Abdelrahman and Punnappurath, Abhijith and Brown, Michael S.},
title = {Leveraging the Availability of Two Cameras for Illuminant Estimation},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2021},
pages = {6637-6646}
}

Contact

Abdelrahman Abdelhamed - (a.abdelhamed@samsung.com; abdoukamel@gmail.com)