A General-Purpose VLM Can Teach an Astronomy Foundation Model to Better Recognize Galaxy Morphology

Researchers tested state-of-the-art vision-language models on a new galaxy morphology VQA benchmark and found they know useful but imperfect morphology cues.

They then used one VLM as a weak teacher to train Zoobot, an astronomy foundation model, for galaxy shape recognition.

Across two survey settings and multiple label budgets, the VLM-guided approach outperformed training with limited human labels alone.

The result points to a practical, label-efficient way to adapt galaxy classifiers for surveys like LSST and the Nancy Grace Roman Space Telescope.