Strong Gravitational Lens Detection 2.0

This project is about detecting strong gravitational lenses in the Kilo Degree Survey.

Based on

This project is based on the following project: https://github.com/CEnricoP/cnn_strong_lensing & paper: https://arxiv.org/abs/1702.07675

Requirements

Use the following command to install required packages:

pip3 install -r requirements.txt

If you wish to make use of the max-tree segmentation as preprocessing step, then you need to take the following steps:

• Get the zip file "siamxt-master.zip" from: https://github.com/rmsouza01/siamxt
• Transfer this file to the machine that you will use (for example Peregrine)
• Load Python, for example I would do this on Peregrine: (This forces Python to be loaded)
  • module load TensorFlow/2.1.0-fosscuda-2019b-Python-3.7.4
  • module load matplotlib/3.1.1-fosscuda-2019b-Python-3.7.4
  • module load scikit-image/0.16.2-fosscuda-2019b-Python-3.7.4
• Now that python is loaded you can install using: (This installs directly from .zip file)

pip3 install siamxt-master.zip --user.

If succesful, then the parameter "do_max_tree_seg" can be set to True.

How to run

python3 main.py --run=runs/experiment_folder/run.yaml

Input parameters of a run can be set in the following file: runs/experiment_folder/run.yaml.

How to view Results

If a run has completed, then a folder with a name such as: "/Strong_Gravitational_Lens_Detection_2.0/models/09_14_2020_09h_48m_56s_name_of_run/" has been created. If you want to compare models against each other than I recommend creating the following directory structure:

models
  experiment4_learning_rate
    07_17_2020_13h_47m_10s_learning_rate_0001
    07_17_2020_14h_13m_09s_learning_rate_001
    07_19_2020_13h_54m_04s_learning_rate_00001

These three models will be compared against each other by running the following:

python3 compare_results.py

This will take you through a dialog that will guide you in plotting results.

Data

Data can be requested.

The data is split in the following way:

• Train Data 80%
• Validation Data 10%
• Test Data 10%

In this binary classification problem three types of images with dimensions (101,101,1) are used:

• 100000 Sources (Simulated Lensing features as .fits files.)
• 5513 Lenses (An image of a galaxy probalby not showing strong gravitational lensing features.)
• 6083 Negatives (An image identified as not showing strong gravitational lensing features.)

More detailed information will be added later on. At this stage in the project, changes will be frequent.