STAR

Space Taxonomy and Analysis Recognition.

Upload a galaxy image and the model predicts whether it is smooth or featured/disk, then shows the confidence and class probabilities.

Data source

This project uses data from the Galaxy Zoo 2 public releases.

Create the following folders inside data/:

• data/raw/
• data/raw/galaxy_zoo_2/

The raw files should be placed in data/raw/galaxy_zoo_2/ like this:

data/
├── raw/
│   └── galaxy_zoo_2/
│       ├── gz2_hart16.csv
│       ├── gz2_filename_mapping.csv
│       └── images/
│           ├── image_1.jpg
│           ├── image_2.jpg
│           └── ...
└── processed/

Download the raw files from the following sources:

• Morphology table: gz2_hart16.csv.gz from the Galaxy Zoo data site
  https://data.galaxyzoo.org/#section-8

• Image files: images_gz2.zip from the Galaxy Zoo 2 Zenodo release
  https://zenodo.org/records/3565489

• Image mapping file: gz2_filename_mapping.csv, used to match image filenames with galaxy entries in the morphology table
  https://zenodo.org/records/3565489/files/gz2_filename_mapping.csv?download=1

After downloading:

• Place gz2_hart16.csv. in data/raw/galaxy_zoo_2/
• Place gz2_filename_mapping.csv in data/raw/galaxy_zoo_2/
• Extract images_gz2.zip
• Move the extracted images/ folder into data/raw/galaxy_zoo_2/

The processed files used during training are saved in data/processed/.

The Galaxy Zoo 2 images are based on galaxies from the Sloan Digital Sky Survey (SDSS).

How it works

The app loads a trained image classification model and processes an uploaded galaxy image before making a prediction.
After inference, it returns the predicted class, the confidence score, and the probability for each class.

Current classes:

• Smooth
• Featured/Disk

How to install

1. Clone the repository:

   git clone https://github.com/Charlsz/star.git
   cd star

2. Create and activate a virtual environment:

   Windows

   python -m venv .venv
   .venv\Scripts\activate

   macOS / Linux

   python -m venv .venv
   source .venv/bin/activate

3. Install the dependencies:

   pip install -r requirements.txt

How to train

Before running the app, you need to train the model to generate the saved weights file used for prediction.

Run:

python star/train.py

After training, the model weights will be saved as:

star_cnn.pt

How to run

After the model has been trained, start the Streamlit app with:

python -m streamlit run star/app.py

Then open the local URL shown in the terminal, upload a galaxy image, and click Predict.