Recurent Color Constancy

Y Qian, K Chen, J Nikkanen, JK Kamarainen, J Matas

ICCV 2017

A Benchmark of Burst Color Constancy

Y Qian, J Käpylä, JK Kämäräinen, S Koskinen, J Matas

ECCV-W 2020

This implementation uses Pytorch.

Installation

Please install Anaconda firstly.

git clone https://github.com/yanlinqian/Temporal-Color-Constancy.git
cd Temporal-Color-Constancy
## Create python env with relevant packages
conda create --name Temporal-Color-Constancy python=3.6
source activate Temporal-Color-Constancy
pip install -U pip
pip install -r requirements.txt
conda install pytorch torchvision cudatoolkit=9.0 -c pytorch  # cudatoolkit=10.0 for cuda10

Tested on pytorch >= 1.0 and python3.

Download

Dataset

Burst Color Constancy Dataset:

• Download the all zip files from the website and unzip them, you will get 400 folders (sequences) in train folder, 200 in test folder.
• Move the train folder and test folder into ./data/, without creating subfolders.
• Run img2npy_temporal.py to convert all pngs to npy files, which will be used to feed the bcc-net.

Run code

Open the visdom service

python -m visdom.server -p 8008

Training

• Train the rcc-net:

python ./rcc_net/train_rccnet.sh

• Train the bcc-net

python ./rcc_net/train_bccnet.sh

Testing

• To reproduce the results reported in the paper, move the pretrained models to ./trained_models/, and then test model directly.

python ./test/test_rccnet.py --pth_path0 ./trained_models/rccnet/fold0.pth
python ./test/test_bccnet.py --pth_path0 ./trained_models/bccnet/fold0.pth