# Multimodal Universe: Galaxy10 DECaLS Galaxy Morphology Images (17,736 Galaxies, HATS)

## Facts

- Provider: Multimodal Universe Collaboration (Polymathic AI, UniverseTBD); Galaxy Zoo DECaLS and DESI Legacy Imaging Surveys
- License: CC-BY-4.0 (https://zenodo.org/api/records/4196267)
- Size: 932 files
- Format: PNG, Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/UniverseTBD/mmu_gz10
- Access: public, readable without an account
- Category: Astronomy & Space
- Source: https://arxiv.org/abs/2412.02527
- Page: https://scigantic.com/archive/c9c79208-fce2-584c-9912-466b61d8377d
- Keywords: Astronomy & Space, benchmark, PNG, Parquet, Hugging Face

## About

Galaxy10 DECaLS, a 10-class galaxy morphology benchmark: RGB image cutouts from the DESI Legacy Imaging Surveys (DECaLS) labelled with confident volunteer classifications from Galaxy Zoo DECaLS, packaged as part of the Multimodal Universe collection. The modality is imaging plus labels: each of 17,736 galaxies carries a 3 by 256 by 256 RGB composite image (the same view Galaxy Zoo volunteers classified), an integer gz10_label over the ten classes (disturbed, merging, round smooth, in-between round smooth, cigar-shaped smooth, barred spiral, unbarred tight spiral, unbarred loose spiral, edge-on without bulge, edge-on with bulge), redshift, pixel scale, ra, dec and object_id. Images are visual composites, not calibrated flux. The README lists 7 columns in 766 HEALPix partitions and 2.6 GiB on disk; about 2.7 GB stored (measured 2026-11-16), main catalog mmu_gz10/ and margin mmu_gz10_10arcs/. Data are a HATS catalog: Apache Parquet partitioned by HEALPix tile, with a 10 arcsecond margin catalog for cross-matching. Open with LSDB (lsdb.open_catalog on the Hugging Face URL, with cone searches and crossmatch), or read the Parquet with pandas, pyarrow, dask or DuckDB. Used to benchmark CNN, vision transformer and self-supervised galaxy morphology classifiers (see the astroNN Galaxy10 page). Galaxy Zoo DECaLS labels are released under CC-BY-4.0 and Legacy Surveys images under CC-BY-4.0; cite Walmsley et al. 2022, Leung and Bovy (astroNN) and the Multimodal Universe paper.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("UniverseTBD/mmu_gz10")
print(ds)
```

## Data structure

File types in sampled files: PNG, Parquet

Example file: `mmu_gz10/dataset/Norder=10/Dir=1150000/Npix=1159494.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/c9c79208-fce2-584c-9912-466b61d8377d
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/c9c79208-fce2-584c-9912-466b61d8377d
- Over MCP: `get_schema_card` with id `c9c79208-fce2-584c-9912-466b61d8377d` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Benchmark datasets](https://scigantic.com/topics/benchmark)
- [Astronomy & Space datasets](https://scigantic.com/topics/astronomy-and-space)
- [PNG datasets](https://scigantic.com/topics/png)
