README.md

---
license: mit
pretty_name: Galaxy Image Description Benchmark
tags:
- astronomy
- galaxy-zoo
- galaxy-morphology
- benchmark
- semantic-search
task_categories:
- image-to-text
dataset_info:
  features:
  - name: image_rgb
    dtype: image
  - name: ra
    dtype: float64
  - name: dec
    dtype: float64
  - name: object_id
    dtype: string
  - name: image_array
    list:
      list:
        list: float32
  - name: decision_tree
    dtype: string
  - name: iauname
    dtype: string
  - name: redshift
    dtype: float64
  - name: elpetro_absmag_r
    dtype: float64
  - name: sersic_nmgy_r
    dtype: float64
  - name: petro_th50
    dtype: float64
  - name: petro_th90
    dtype: float64
  - name: petro_theta
    dtype: float64
  - name: upload_group
    dtype: string
  - name: active_learning_on
    dtype: bool
  - name: in_gzd_ab
    dtype: bool
  - name: png_loc
    dtype: string
  - name: smooth-or-featured_total-votes
    dtype: int64
  - name: smooth-or-featured_smooth
    dtype: int64
  - name: smooth-or-featured_smooth_fraction
    dtype: float64
  - name: smooth-or-featured_featured-or-disk
    dtype: int64
  - name: smooth-or-featured_featured-or-disk_fraction
    dtype: float64
  - name: smooth-or-featured_artifact
    dtype: int64
  - name: smooth-or-featured_artifact_fraction
    dtype: float64
  - name: disk-edge-on_total-votes
    dtype: int64
  - name: disk-edge-on_yes
    dtype: int64
  - name: disk-edge-on_yes_fraction
    dtype: float64
  - name: disk-edge-on_no
    dtype: int64
  - name: disk-edge-on_no_fraction
    dtype: float64
  - name: has-spiral-arms_total-votes
    dtype: int64
  - name: has-spiral-arms_yes
    dtype: int64
  - name: has-spiral-arms_yes_fraction
    dtype: float64
  - name: has-spiral-arms_no
    dtype: int64
  - name: has-spiral-arms_no_fraction
    dtype: float64
  - name: bar_total-votes
    dtype: int64
  - name: bar_strong
    dtype: int64
  - name: bar_strong_fraction
    dtype: float64
  - name: bar_weak
    dtype: int64
  - name: bar_weak_fraction
    dtype: float64
  - name: bar_no
    dtype: int64
  - name: bar_no_fraction
    dtype: float64
  - name: bulge-size_total-votes
    dtype: int64
  - name: bulge-size_dominant
    dtype: int64
  - name: bulge-size_dominant_fraction
    dtype: float64
  - name: bulge-size_large
    dtype: int64
  - name: bulge-size_large_fraction
    dtype: float64
  - name: bulge-size_moderate
    dtype: int64
  - name: bulge-size_moderate_fraction
    dtype: float64
  - name: bulge-size_small
    dtype: int64
  - name: bulge-size_small_fraction
    dtype: float64
  - name: bulge-size_none
    dtype: int64
  - name: bulge-size_none_fraction
    dtype: float64
  - name: how-rounded_total-votes
    dtype: int64
  - name: how-rounded_round
    dtype: int64
  - name: how-rounded_round_fraction
    dtype: float64
  - name: how-rounded_in-between
    dtype: int64
  - name: how-rounded_in-between_fraction
    dtype: float64
  - name: how-rounded_cigar-shaped
    dtype: int64
  - name: how-rounded_cigar-shaped_fraction
    dtype: float64
  - name: edge-on-bulge_total-votes
    dtype: int64
  - name: edge-on-bulge_boxy
    dtype: int64
  - name: edge-on-bulge_boxy_fraction
    dtype: float64
  - name: edge-on-bulge_none
    dtype: int64
  - name: edge-on-bulge_none_fraction
    dtype: float64
  - name: edge-on-bulge_rounded
    dtype: int64
  - name: edge-on-bulge_rounded_fraction
    dtype: float64
  - name: spiral-winding_total-votes
    dtype: int64
  - name: spiral-winding_tight
    dtype: int64
  - name: spiral-winding_tight_fraction
    dtype: float64
  - name: spiral-winding_medium
    dtype: int64
  - name: spiral-winding_medium_fraction
    dtype: float64
  - name: spiral-winding_loose
    dtype: int64
  - name: spiral-winding_loose_fraction
    dtype: float64
  - name: spiral-arm-count_total-votes
    dtype: int64
  - name: spiral-arm-count_1
    dtype: int64
  - name: spiral-arm-count_1_fraction
    dtype: float64
  - name: spiral-arm-count_2
    dtype: int64
  - name: spiral-arm-count_2_fraction
    dtype: float64
  - name: spiral-arm-count_3
    dtype: int64
  - name: spiral-arm-count_3_fraction
    dtype: float64
  - name: spiral-arm-count_4
    dtype: int64
  - name: spiral-arm-count_4_fraction
    dtype: float64
  - name: spiral-arm-count_more-than-4
    dtype: int64
  - name: spiral-arm-count_more-than-4_fraction
    dtype: float64
  - name: spiral-arm-count_cant-tell
    dtype: int64
  - name: spiral-arm-count_cant-tell_fraction
    dtype: float64
  - name: merging_total-votes
    dtype: int64
  - name: merging_none
    dtype: int64
  - name: merging_none_fraction
    dtype: float64
  - name: merging_minor-disturbance
    dtype: int64
  - name: merging_minor-disturbance_fraction
    dtype: float64
  - name: merging_major-disturbance
    dtype: int64
  - name: merging_major-disturbance_fraction
    dtype: float64
  - name: merging_merger
    dtype: int64
  - name: merging_merger_fraction
    dtype: float64
  - name: wrong_size_statistic
    dtype: float64
  - name: wrong_size_warning
    dtype: bool
datasets:
- legacy
---

# Description Benchmark based on GalaxyZoo
[![arXiv](https://img.shields.io/badge/arXiv-2512.11982-b31b1b.svg)](https://arxiv.org/abs/2512.11982)
[Paper](https://huggingface.co/papers/2512.11982) | [Project Page](https://aion-search.github.io) | [Code](https://github.com/NolanKoblischke/AION-Search)

A small sample of the GZ-DECaLS dataset to cheaply test the ability of vision-language models to describe galaxy images.

## Sample Usage

```python
from aionsearch import AIONSearchClipModel

# Load pretrained model from HuggingFace
model = AIONSearchClipModel.from_pretrained()

# Project AION image embeddings into shared space
aion_embedding = # Embedding of an image using github.com/PolymathicAI/AION
projected_image = model.image_projector(aion_embedding)  # (batch, 768) -> (batch, 1024)

# Project OpenAI text embeddings into shared space  
text_embedding = # Embedding of text using text-embedding-3-large
projected_text = model.text_projector(text_embedding)    # (batch, 3072) -> (batch, 1024)

# Compute similarity for semantic search
similarity = projected_image @ projected_text.T
```
See [\`examples/quick_start.ipynb\`](https://github.com/NolanKoblischke/AION-Search/blob/main/examples/quick_start.ipynb) for a complete walkthrough that downloads a galaxy image, generates embeddings with AION, and performs text-to-image similarity search.

## Data source
The underlying catalog data are derived from the Galaxy Zoo DECaLS: Detailed Visual Morphology Measurements from Volunteers and Deep Learning for 314,000 Galaxies and remain subject to their original data-use policy and required acknowledgements.

## Embeddings Citation
```bibtex
@misc{koblischke2025semantic,
      title={Semantic search for 100M+ galaxy images using AI-generated captions}, 
      author={Nolan Koblischke and Liam Parker and Francois Lanusse and Irina Espejo Morales and Jo Bovy and Shirley Ho},
      year={2025},
      eprint={2512.11982},
      archivePrefix={arXiv},
      primaryClass={astro-ph.IM},
      url={https://arxiv.org/abs/2512.11982}, 
}
```

## GZ-DECaLS Dataset Citation
```bibtex
@dataset{walmsley_mike_2020_4573248,
  author       = {Walmsley, Mike and
                  Lintott, Chris and
                  Tobias, Geron and
                  Kruk, Sandor J and
                  Krawczyk, Coleman and
                  Willett, Kyle and
                  Bamford, Steven and
                  Kelvin, Lee S and
                  Fortson, Lucy and
                  Gal, Yarin and
                  Keel, William and
                  Masters, Karen and
                  Mehta, Vihang and
                  Simmons, Brooke and
                  Smethurst, Rebecca J and
                  Smith, Lewis and
                  Baeten, Elisabeth M L and
                  Macmillan, Christine},
  title        = {{Galaxy Zoo DECaLS: Detailed Visual Morphology 
                   Measurements from Volunteers and Deep Learning for
                   314,000 Galaxies}},
  month        = dec,
  year         = 2020,
  publisher    = {Zenodo},
  version      = {0.0.2},
  doi          = {10.5281/zenodo.4573248},
  url          = {https://doi.org/10.5281/zenodo.4573248}
}