A Generative Model for Disentangling Galaxy Photometric Parameters

TL;DR

This work tackles the scalability challenge of extracting galaxy morphological parameters from vast photometric surveys by introducing a conditional autoencoder trained on GalSim-generated mocks. The CAE learns a disentangled latent space where a small, physically supervised subset encodes flux and half-light radius while the rest captures residual morphology, enabling accurate image reconstruction in a single forward pass. Compared with PCA, the CAE shows superior ability to capture nonlinear morphology–parameter relationships, achieving higher $R^2$ values with far fewer latent dimensions. The approach holds promise for fast, scalable morphology analysis in future surveys, though real-data domain adaptation and extension to additional parameters remain important avenues for development.

Abstract

Ongoing and future photometric surveys will produce unprecedented volumes of galaxy images, necessitating robust, efficient methods for deriving galaxy morphological parameters at scale. Traditional approaches, such as parametric light-profile fitting, offer valuable insights but become computationally prohibitive when applied to billions of sources. In this work, we propose a Conditional AutoEncoder (CAE) framework to simultaneously model and characterize galaxy morphology. Our CAE is trained on a suite of realistic mock galaxy images generated via GalSim, encompassing a broad range of galaxy types, photometric parameters (e.g., flux, half-light radius, Sersic index, ellipticity), and observational conditions. By encoding each galaxy image into a low-dimensional latent representation conditioned on key parameters, our model effectively recovers these morphological features in a disentangled manner, while also reconstructing the original image. The results demonstrate that the CAE approach can accurately and efficiently infer complex structural properties, offering a powerful alternative to existing methods.

Figures (7)

• Figure 1: Representation of the conditional autoencoder, taking an input image of a galaxy to process through the encoder, latent space and a decoder to reconstruct the output image for photometric analysis. The latent space is supervised to learn the physical parameters of the galaxy image. Detailed parameters can be found in the Appendix.
• Figure 2: Sample reconstructions from the autoencoder. Top row: input galaxy images. Bottom row: corresponding reconstructions, normalized to the same range as the input on top.
• Figure 3: Correlation between physical parameters and latent dimensions. Left: flux vs $z_1$; right: half-light radius vs $z_2$. Each point represents a galaxy image sample.
• Figure 4: Comparison of $R^2$ between input and output parameters for train and test datasets.
• Figure 5: Effect of varying individual latent dimensions. Top row: increasing $z_1$ (flux) causes brightness to increase while structure remains fixed. Bottom row: increasing $z_2$ (half-light radius) leads to a visibly larger galaxy size. All other latent variables are held constant; both rows originate from the same base galaxy.