astromorph performs automatic, self-supervised morphological classification of astronomical objects from FITS images using a Python machine learning pipeline. It implements the BYOL (Boot-strap Your Own Latents) framework to learn embeddings without labeled training data, providing a lightweight 2D convolutional network, dataset utilities for FITS filelists, and configurable training and inference scripts. astromorph separates objects by morphology through learned representations, exports embeddings and metadata for downstream analysis, and supports both flexible PyTorch-level use and easier command-line workflows for astronomy image datasets.