A Guided Unconditional Diffusion Model to Synthesize and Inpaint Radio Galaxies from FIRST, MGCLS and Radio Zoo

TL;DR

The growing data volume from next-generation radio surveys makes manual analysis infeasible, motivating synthetic data generation and inpainting for radio galaxies. The authors employ a masked guided diffusion model trained on a multi-survey dataset (FIRST, MGCLS, Radio Galaxy Zoo) to synthesize and inpaint radio-galaxy images without retraining, using a U-Net–style architecture and a training regime that blends partially masked and fully masked samples. Generated images visually resemble real observations and closely match several statistical properties of real data, including pixel-value distributions and basic source characteristics, though challenges remain for extreme brightness and source clustering. This approach provides a practical data augmentation and restoration tool for ML workflows in radio astronomy, supporting robust analysis for large-scale surveys such as the SKA.

Abstract

We present a masked guided approach for a denoising diffusion probabilistic model (DDPM) trained to generate and inpaint realistic radio galaxy images. We train the DDPM using the FIRST radio galaxy catalog, the Radio Galaxies Zoo and cutouts of the MGCLS catalog. We compared different statistical distributions to make sure that our unconditional approach produces morphologically realistic galaxies, offering a data-driven method to supplement existing radio datasets and support the development of machine learning applications in radio astronomy.

Figures (15)

• Figure 1: Illustration of the forward and backward process in a DDPM. The forward process (blue arrows) progressively adds Gaussian noise to an initial radio galaxy image $\mathrm{x_0}$, transforming it into a pure noise distribution $\mathrm{x_T}$ over $\mathrm{T}$ time steps. The backward process (red arrows) aims to learn the denoising operation $\mathrm{p_\theta(x_t \mid x_{t-1})}$, gradually removing noise to reconstruct the original image.
• Figure 2: Example of some masked images used for training. The masked regions are represented in gray
• Figure 3: Example of real images used to train our trained DDPM.
• Figure 4: Example of images generated by our DDPM.
• Figure 5: Statistical comparison of the brightest pixel values in each image in the real data sample and the generated one by our network