# CSST-COMMOM-CRMASK - Introduction

**CRMask** is a module for detecting cosmic rays on the CSST images using state-of-art models. It should be also applied for images from other telescopes, as long as the images have the same data structure as that of CSST.

# Python package dependencies

The following python packages are needed to **CRMask**:

* astropy == 4.x (>= 5.0 will be tested in the future) 
* ccdproc >= 2.1.0
* deepCR >= 0.1.5
* matplotlib >= 3.1.2
* numpy >= 1.19.5
* pytorch >= 1.8.1
* scipy >= 1.3.3

Install these package under Ubuntu 20.04 LTS:

```
sudo apt install python3-numpy python3-scipy python3-matplotlib python3-astropy python3-ccdproc
sudo python3 -m pip install pytorch deepCR
```
#File list

* crmask.py, cosmic ray detection class, it can be either used standalone or imported as a Python module.
* crmask.ini, configuration file that will be loaded when CRMask class initialized.
* CSST\_2021-12-30\_CCD23\_epoch20.pth, training models obtained from CSST simulated iimages with 50 hidden layers and 50 epoch.  

# Usage

Download crmask.py and crmask.ini, put them in the same directory.

Running it:

```
./crmask.py your_file.fits your_model
```
where **your\_file.fits** is the filename of the image which you want to do cosmic ray detection, and **your\_model** is the algorithm you select. Available algorithms are _deepCR_ and _lacosmic_.

<!--
下载crmask.py，crmask.ini这两个文件。

./crmask.py your_file.fits your_model

其中your_file.fits是输入fits文件，your_model是扣除宇宙线的方法，可以是deepCR或者lacosmic。

如果需要使用CSST仿真数据训练得到的模型扣除宇宙线，下载CSST_2021-12-30_CCD23_epoch20.pth，该模型隐藏层数为50。
-->

# Warning

* The default training model of deepCR are obtained from HST/ACS data by Zhang and Bloom (2020). If you would like to use the native model rather than the model **CSST\_2021-12-30\_CCD23\_epoch20.pth** obtained from CSST simulated data by **Hu Yi**, you must also change the **hidden** parameters in the configuration file **crmask.ini** to 32. 

* The input images must be **bias-substracted** (either by bias reference fram or overscan region) no matter which algorithm selected, otherwise you will get very inefficient detection rate thus wrong results. Flat-field corrected images are preferable, but not necessary. 
<!--
If you would like use the native training model of deepCR, which are obtained from HST/ACS
如果使用deepCR自带用HST/ACS数据训练得到的模型，在配置文件中，要将hidden这个配置参数改为32。-->

# Bugs

Please contact **_HU Yi_** (huyi.naoc@gmail.com) to report bugs.