This repository accompanies the paper: deepCR: Cosmic Ray Rejection with Deep Learning (Zhang & Bloom (2019)), and includes code to reproduce results (figures and tables) of the paper.
deepCR is implemented separately in: https://github.com/profjsb/deepCR.
Tested to work on Python 3.6 and 3.7
Automatically runs on GPU if torch.cuda.is_available()
pip install -r requirements_pip.txt
cd paper/data/
sh generate_data.sh
cd ../
sh run_all.sh
Figures and tables are by default generated from pre-calculated benchmarking data saved in paper/benchmark_data/*.npy files
If you would like to reproduce benchmarking results from scratch, simply delete these *.npy files.
Warning: it is highly recommended that benchmarking be run on GPU(s). On CPUs they're expected to run for hours.