nidhihegde001 / crismis

Cosmic-Ray Imaging Studies via Mission-Imagery from Space

CRISMIS - Google Summer of Code 2020 (CERN)

CRISMIS is an open-source machine learning based tool to identify cosmic-ray artifacts in imaging data. Its purpose is to cater to the needs of planetary scientists and space science teams and help them to automatically filter, and search for the presence of cosmic ray artefacts in their data.

In addition to this, it also provides the flexibility to train on your own data.

Installation

---

1. Clone this repo

2. Install the following packages in your environment:

   • Pytorch 1.6.0, cudatoolkit 10.1, torchvision (refer : https://pytorch.org/)
   • planetaryimage 0.5.0 pip install planetaryimage
   • requests 2.7 pip install requests
   • Astropy 4.0 pip install astropy, Scipy 1.4 pip install scipy
   • Matplotlib python -m pip install -U matplotlib
   • Tqdm pip install tqdm

3. To use the tool, download the 2 pre-trained models rpn_99.pth and classifier_final.pth and place it in the main directory of this repo. The link to the models is:

   • https://drive.google.com/drive/folders/18ciTnhn0-aSIqPd5NrWT6mfCpEI0Nat4?usp=sharing

Usage:

---

Tool:

1. To use the tool, first follow the Installation steps above.
2. Type python tool.py with the required arguments
   • --directory name of the directory
   • --name name of the image in directory
3. A typical example looks like this:

    python tool.py --directory 2014_215 --name EN1049375684M.IMG

The result looks like this:

Every tested image is stored in the images/ subdirectory and its result is stored in the directoryprediction/, if a cosmic ray is found. If the cosmic ray is not found, the message Cosmic Ray not Found is displayed.

Training Script:

1. In case you want to train the RPN model yourself on a different dataset, you may use the training script provided in the directory rpn/
2. Go to the rpn subfolder
3. Type python train.py with the required arguments
   • --e number of epochs to train
   • --model Backbone : resnet50 or resnet18 are provided as of now
   • --exp Experiment No if you want to perform any experiments
4. A typical example for training looks like this:

     python train.py --e 10 --model resnet50 --exp 1

5. The training script saves the model after every 20 epochs in the directory saved_models/exp_no where 'exp_no' is the experiment number.
6. Along with the saved models, 3 graphs are generated corresponding to classification, regression and total losses during training and validation.
7. For visualization of the trained model, the script visualize.py is provided.

Notebooks

The work has also been distributed to the following notebooks for easy use:

crismis_tool.ipynb This is a colab version of tool, provided you do not want to go through the hassle of creating environments on your local machine. (Colab comes pre-installed with most of the libraries)

scanner.ipynb Contains utilities for easy retrieval of data from the MESSENGER archive, a GUI for sorting the images based on visual inspection, and a scanner to scan the images for cosmic ray artefacts from a list of dates from the archive.

artefacts_library.ipynb Pre-processing of predictions at the rpn stage to help in creation of different classes of data for the classifier stage

RPN_R50_notebook.ipynb The training script contained in the rpn/ folder has been converted into the form of a notebook, for step by step training and visualisation.

classifier_Net.ipynb A notebook containing the implementation of the classifier trained on crops of different artifacts

crismis's Issues

UnboundLocalError: local variable 'a' referenced before assignment

When I cloned this repo and try to use this tool It was showing me this error UnboundLocalError: local variable 'a' referenced before assignment.

This is the screenshot of the error that I got

can anyone let me know how to fix this error??