msiudek/astroPT_euclid_VIS_NISP_SED_embeddings

---
dataset_info:
features:
- name: object_id
dtype: int64
- name: embedding
sequence: float32
splits:
- name: train
num_bytes: 816643200
num_examples: 264800
- name: test
num_bytes: 206504640
num_examples: 66960
download_size: 1229793272
dataset_size: 1023147840
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
---

# AstroPT Euclid Embeddings

Pre-computed embeddings from the [AstroPT VIS+NISP+SED Model](https://huggingface.co/msiudek/astroPT_euclid_VIS_NISP_SED_model) on Euclid dataset samples.

## Overview

This repository contains pre-computed feature embeddings generated by AstroPT models applied to the Euclid Q1 galaxy dataset. These embeddings can be used for

- Efficient downstream task training (reduced computational cost)
- Feature analysis and visualization
- Similarity search and retrieval
- Clustering and unsupervised learning
- Fast fine-tuning on specialized tasks

Example notebooks are found in the [AstroPT scripts](https://github.com/Smith42/astroPT/tree/main/scripts/euclid)

Available embeddings:
- **VIS**: Single-band embeddings from Euclid VIS imaging; [VIS embeddings](https://huggingface.co/msiudek/astroPT_euclid_VIS_embeddings)
- **VIS+NISP**: Multi-band embeddings from VIS + 3× NIR (Y, J, H); [VIS+NISP embeddings](https://huggingface.co/msiudek/astroPT_euclid_VIS_NISP_embeddings)
- **VIS+NISP+SED**: Multi-modal embeddings from imaging + 13-band photometry; [VIS+NISP+SED embeddings](https://huggingface.co/msiudek/astroPT_euclid_VIS_NISP_SED_embeddings)

## Quick Start

### Load VIS+NSIP Embeddings

```python
from datasets import load_dataset

# Load VIS embeddings
embeddings = load_dataset(
"msiudek/astroPT_euclid_VIS_NISP_SED_embeddings",
split="train",
streaming=True
)

# View a sample
sample = embeddings[0]
print(f"Object ID: {sample['object_id']}")
print(f"Embedding shape: {len(sample['embedding'])}")
print(f"Embedding: {sample['embedding']}")
```

## Related Resources

**Models**:
- [AstroPT VIS Model](https://huggingface.co/msiudek/astroPT_euclid_VIS_model)
- [AstroPT VIS+NISP Model](https://huggingface.co/msiudek/astroPT_euclid_VIS_NISP_model)
- [AstroPT VIS+NISP+SED Model](https://huggingface.co/msiudek/astroPT_euclid_VIS_NISP_SED_model)

**Datasets**:
- [AstroPT Euclid Dataset](https://huggingface.co/datasets/msiudek/astroPT_euclid_dataset): Original imaging + photometry
- [AstroPT Euclid Metadata](https://huggingface.co/datasets/msiudek/astroPT_euclid_metadata): Galaxy properties

**Code**:
- [AstroPT GitHub](https://github.com/Smith42/astroPT): Training and inference code

## Citation

```bibtex
@article{Siudek2025,
title={AstroPT: Astronomical Physics Transformers for Multi-modal Learning},
author={Siudek, M and others},
journal={Euclid Collaboration},
eprint={2503.15312},
archivePrefix={arXiv},
year={2025},
url={https://ui.adsabs.harvard.edu/abs/2025arXiv250315312E/abstract}
}
```

## License

CC-BY-4.0

---

**Last Updated**: December 2025
**Embeddings Version**: 1.0