Group repo for the tidal CNN project.
Make sure you are using a python environment that you are comfortable changing. Always good practice to make a new virtual environment or conda environment for a new project. If it requires activation, activate it.
I use Song Huang’s unagi package — which you will have to clone from its git repository:
git clone https://github.com/dr-guangtou/unagi.git
You must then go into the unagi directory and do:
python setup.py install
This will put an unagi egg in the site-packages directory of your python module library located at path_to_my_env/lib/pythonX.X/site-packages. You can check the installation by moving to a directory that does not contain the unagi directory and starting python:
import unagi
If that does not throw an error, your installation is stable.
In order to download images, you must have a Subaru account: https://hsc-release.mtk.nao.ac.jp/datasearch/new_user/new
You will select a username and be given a (long) password. Once you have a username and password, add them as environment variables to your .bash_profile or .bashrc or .profile. Alternatively, you can also add them to the activate file for your python environment. These steps prevent you from being asked for your username and password for every download.
In bash:
export SSP_PDR_USR='xxx'
export SSP_PDR_PWD='yyy'
I have written a short script, GetHSC_Cutouts.py, for downloading the images. Modify it however you like on your branch. It currently grabs images in all grizy bands, making cutouts roughly 525x525 in size. I picked this size because the WCS matrices are not the same for every location accross the survey... Meaning that some images will be 524x526, etc. I can trim all of these down to 512x512 in post-processing, then resample to lower resolution or cutout a specified region.
Go into the GetHSC_Cutouts.py file and change the path to the Catalogues and your desired output directory (which I have confusingly called Input in my folder since I consider it as input for CNNs. You will then be ready to download images. Message me on slack if you have an issue.