Foundation Models in Astronomy: Why They Matter

Mon, 10 Feb 2025 00:00:00 +0000

Foundation models — large neural networks pre-trained on massive datasets — are starting to transform astronomical data analysis. Models like DINOv2 or CLIP, originally trained on natural images, can be fine-tuned for astronomical tasks with surprisingly good results. At , I presented early comparisons of these models on galaxy morphological classification, and my paper digs into how to best specialize DINOv2 for astronomy.

The key challenge is the domain gap: astronomical images (multi-band, high dynamic range, specific noise) look nothing like everyday photos. Choosing the right fine-tuning strategy turns out to matter a lot — and is the core focus of my current work.

On the multimodal side, I am involved in the collaboration, where I contribute to — a vision-language model for astronomy presented at ICLR 2025.

Galaxy Morphological Classification with Deep Learning

Fri, 15 Nov 2024 00:00:00 +0000

Galaxy morphological classification — spiral, elliptical, irregular? — is one of the most natural applications of computer vision in astronomy. Datasets like and provide crowd-sourced morphological labels for thousands of galaxies and serve as standard benchmarks.

In my work with Sébastien Lefèvre, we compared several pre-trained foundation models (ViT, SwinV2, BEiT, DINOv2) on Galaxy10 DECaLS to see how well they transfer to astronomical images. The short answer: surprisingly well, but the fine-tuning strategy matters a lot. These results were presented at and , with a detailed analysis in the .

Deep Learning |

Foundation Models in Astronomy: Why They Matter

Galaxy Morphological Classification with Deep Learning