Galaxy Morphology

Machine Learning

Python

HTML

Deep learning for deep space: Classifying JWST galaxy morphologies with Convolutional Vision Transformers.

Published

April 23, 2026

Project Overview

This is an interactive visualization of learned feature representations from a CvT-13 model fine-tuned on Galaxy Zoo 2 images. The target data here are unlabeled JWST Advanced Early Extragalactic Survey (JADES) cutouts generated by stacked NIRCam filters and segmentation mask available at: https://archive.stsci.edu/hlsp/jades. Explore how the model clusters different galaxy morphologies (round, in-between, cigar-shaped, edge-on, spiral) in embedding space, with each point representing a galaxy’s compressed visual features. The remaining unconfident set of images are left for expert review to filter point sources or discover additional morphology classes.