This dataset accompanies the LSBGPENet paper on automated structural and photometric parameter estimation for low-surface-brightness galaxies (LSBGs) in the Dark Energy Survey (DES). It contains two independent samples. For each sample, we provide RGB composite JPEG images and one matching parameter table (CSV) in separate files.
All images were retrieved from DES DR1 tiles via the NOIRLab Astro Data Lab (DECam on the Blanco 4 m telescope; optical g, r, i bands). Images are 64 × 64 pixels (16.832″ × 16.832″; pixel scale 0.263 arcsec pixel⁻¹) with pixel-by-pixel Galactic extinction correction (SFD98). RGB composite JPEG images are Lupton et al. (2004) color composites.

Sample 1: Tanoglidis (2021) — 23,111 objects
Images (per object): one RGB JPEG.
Parameters: catalog_23111.csv — training labels derived with GALFITM multi-band Sérsic fitting (Tanoglidis et al. 2021, ApJS), used as ground-truth values for LSBGPENet training, validation, and testing (8:1:1 split: 18,489 / 2,311 / 2,311). One CSV row per object, matched to images by object ID.

Sample 2: Thuruthipilly (2024) — 3,728 objects
Images (per object): one RGB JPEG (same processing as Sample 1).
Parameters: catalog_3728_LSBGPENet.csv — LSBGPENet-predicted structural and photometric parameters with associated uncertainties (ellipticity, Sérsic index n, g/r/i effective radii and magnitudes, mean and central surface brightnesses; column definitions as in Table catalog_desc of the paper). One CSV row per object, matched to images by object ID.