deepCR-ACS/WFC
To reproduce the models, run deepCR-ACS-WFC.py in the command line. It will sequentially execute
-
Downloading the training and test datasets
-
Fetching the training data and test directories.
-
Training and testing the models. You'll need scalable_datasets branch set up for this step.
- The models that will be trained are ACS/WFC F435W, F606W, F814W individual filter models, and all-in-one global model.
- The individual filter models are tested with respective test dataset, and the global model is tested with all three test datsets concatenated. Additionally, LACosmic can be tested with the individual test datasets if desired.
-
Plotting the test results. (ROC curves)
You'll likely need ~50GB disk space to download the training and test data, as well as GPU support to intiate training.
Please cite the following if you use deepCR on ACS/WFC in your paper
@article{2021RNAAS...5...98K,
author = {Kwon, K.~J. and Zhang, Keming and Bloom, Joshua S.},
doi = {10.3847/2515-5172/abf6c8},
eid = {98},
journal = {Research Notes of the American Astronomical Society},
keywords = {Astronomy data reduction, Convolutional neural networks, Classification, Neural networks, Cosmic rays, Hubble Space Telescope, Astronomical detectors, 1861, 1938, 1907, 1933, 329, 761, 84},
month = {April},
number = {4},
pages = {98},
title = {DEEPCR on ACS/WFC: Cosmic-Ray Rejection for HST ACS/WFC Photometry},
volume = {5},
year = {2021}
}