The Multi-Band Galaxy Benchmark: Human-Labeled Properties from Radio to X-Ray
JIA PENG
The AstroM³ (Astro Multi-modal, Multi-band, Multi-semantic) dataset is a pioneering data infrastructure designed to bridge the "modality gap" in extragalactic astrophysics. While galaxies are complex, multi-component laboratories of cosmic evolution, current astronomical resources remain fragmented across isolated surveys. AstroM³ addresses this by unifying panchromatic imaging with qualitative insights from scientific literature, creating a standardized framework that enables the development of interpretable, multi-modal artificial intelligence for the first time.The dataset provides a comprehensive panchromatic view of approximately 6 million galaxies, integrating data across 25 electromagnetic bands. This range spans from high-energy X-ray (Chandra) and ultraviolet (GALEX) to optical (DESI, J-PLUS), mid-infrared (WISE), and radio (VLA) observations. With spatial resolutions reaching $0.262\,\mathrm{arcsec}$ and optical spectral resolutions of $R \approx 2000\text{--}2300$, the dataset aligns high-fidelity imaging with spectroscopic depth, ensuring physical consistency across the entire electromagnetic spectrum.Beyond raw observations, AstroM³ incorporates structured semantic metadata that links quantitative physical parameters to the qualitative classifications established in decades of astronomical literature. This "multi-semantic" approach is specifically designed for deep learning workflows, allowing for physics-informed machine learning and "chain-of-thought" prompting. To ensure technical scalability, the data is organized using HEALPix indexing and stored in HDF5 format, facilitating efficient cross-modal querying and high-performance parallel I/O for large-scale research.
Files
Paper Information
Paper Title:
GalaxSem: A Multi-modal and Multi-wavelength Dataset with Structured Semantic Metadata for Interpretable Galaxy Astrophysics
Publication:
Scientific Data
Identifiers
CSTR:
11379.11.101762
DOI:
10.12149/101762
VO Identifier:
ivo://China-VO/paperdata/101762
Publication Date:
2026-08-18
Usage Statistics
Total Downloads
44136
Citations
JIA PENG. The Multi-Band Galaxy Benchmark: Human-Labeled Properties from Radio to X-Ray[DS/OL]. 1.0. National Astronomical Data Center, 2026[2026-09-10]. https://doi.org/10.12149/101762.
BibTeX
@misc{10.12149/101762,
doi = {10.12149/101762},
url = {https://doi.org/10.12149/101762},
author = {JIA PENG},
title = {The Multi-Band Galaxy Benchmark: Human-Labeled Properties from Radio to X-Ray},
version = {1.0},
publisher = {National Astronomical Data Center of China},
year= {2026}
}
Versions
Main
This DOI represents all versions, and will always resolve to the latest one.
2026-08-18