https://github.com/yash-10/jellyfish_self_supervised
Code accompanying "Systematic analysis of jellyfish galaxy candidates in Fornax, Antlia, and Hydra from the S-PLUS survey: a self-supervised visual identification aid"
https://github.com/yash-10/jellyfish_self_supervised
galaxies jellyfish self-supervised-learning
Last synced: 6 months ago
JSON representation
Code accompanying "Systematic analysis of jellyfish galaxy candidates in Fornax, Antlia, and Hydra from the S-PLUS survey: a self-supervised visual identification aid"
- Host: GitHub
- URL: https://github.com/yash-10/jellyfish_self_supervised
- Owner: Yash-10
- License: mit
- Created: 2022-08-28T17:36:01.000Z (about 3 years ago)
- Default Branch: main
- Last Pushed: 2024-11-29T04:43:55.000Z (10 months ago)
- Last Synced: 2025-02-04T19:29:28.602Z (8 months ago)
- Topics: galaxies, jellyfish, self-supervised-learning
- Language: Jupyter Notebook
- Homepage: https://academic.oup.com/mnras/article/532/1/270/7689212
- Size: 8.46 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Jellyfish Galaxies: a self-supervised approach for assisting visual identification
[](https://colab.research.google.com/drive/1I7lIh-vaYhlifvULbzfU3enas008jxre?usp=sharing)
**Motivation**\
*Jellyfish galaxies* are galaxies undergoing transformation. These galaxies serve as an example of extreme cases of a phenomenon called ram pressure stripping (RPS). This phenomenon causes the removal of the cold gas from the gaseous disks of the galaxies, resulting in extended tails of gas. As a result, these galaxies possess peculiar morphologies and properties that distinguish them from other galaxies. These galaxies are rare to find. The traditional approach to detecting jellyfish galaxies is visual inspection, where visual classifiers look at specific characteristics (gas tail, disturbed morphology, etc). Based on the visual strengths of these features, a JClass is assigned to a galaxy, where a JClass of 0 means no evidence of RPS, and a JClass of 5 means the most extreme evidence of RPS.
However, visual analysis is subjective and is affected by human biases. It is also infeasible to manually vet through large astronomical datasets to detect these rare galaxies. This work aims to assist visual classification using self-supervised learning to improve the quality of the visual classification. Specifically, for each visually classified galaxy (yielding a JClass), we assign an equivalent JClass based on self-supervised learning. Our approach alleviates human biases and is scalable to large datasets.
## Scientific details
Please see our paper: ["Systematic analysis of jellyfish galaxy candidates in Fornax, Antlia, and Hydra from the S-PLUS survey: a self-supervised visual identification aid"](https://academic.oup.com/mnras/article/532/1/270/7689212).
## Data
The trained models, catalogs, and related files can be found [here](https://drive.google.com/drive/folders/1HTWDpad8P7trQN_od8FFc6qIdNJ_AfqQ?usp=sharing).
## Repository description
The Colab notebook provides code for how to set up data for training the self-supervised model, presents an example demonstration of the query by example, how the uncertain visual classification of jellyfish galaxies can be improved using self-supervised learning, Grad-CAM analysis, and trail vector analysis of the galaxies. Some miscellaneous code is also present in the notebook.
- `catalogs_analysis` contains catalogs and related codes for generating the plots of the astrophysical analysis.
- `data_utils` contains helper scripts for handling data for the deep learning application.
- `self_supervised` contains code for creating the self-supervised (here, SimCLR) model, evaluation metrics, and other helper functions.
- `old_temp_store` contains codes not used in our analysis but were developed at some point during the work. This folder is kept only for storage purposes.
- `pretrain.py` is the script used to pre-train the self-supervised model.
- `baseline_resnet_supervised.py` contains code for training the supervised CNN model.
- `cross_validate.py` contains code for the K-fold cross-validation for the self-supervised approach, and `cross_validate_linear_eval.py` contains code for the K-fold cross-validation of the linear evaluation protocol (i.e., a linear classifier trained on the self-supervised representations).
**Note**: The procedure for estimating the star formation rates and associated codes is present in the following GitHub repository: https://github.com/amanda-lopes/Halpha-SPLUS-Jelly
## References
The `SimCLR` code in this repository is inspired by the SimCLR tutorial in the UvA DL Notebooks: [Tutorial 17: Self-Supervised Contrastive Learning with SimCLR](https://uvadlc-notebooks.readthedocs.io/en/latest/tutorial_notebooks/tutorial17/SimCLR.html)
## License
[MIT](https://github.com/Yash-10/jellyfish_self_supervised/blob/main/LICENSE)