# AI-ready Euclid VIS Galaxy Cutout Dataset

This dataset provides an AI-ready collection of Euclid VIS galaxy image
cutouts together with curated morphology annotations and precomputed
representation vectors. It is constructed based on the Galaxy Zoo Euclid
(Q1) morphology catalog.

## Dataset Contents

### 1. Image Cutouts (FITS)

Galaxy-centered image cutouts extracted from Euclid VIS observations.

-   Total number of galaxies: 365,513\
-   Format: FITS\
-   Image size: 224 × 224 pixels

Distributed as multiple archives: - euclid_vis_part01.zip -
euclid_vis_part02.zip - euclid_vis_part03.zip - euclid_vis_part04.zip -
euclid_vis_part05.zip

------------------------------------------------------------------------

### 2. Morphology Annotation Catalog

-   morphology_catalogue_with_high_quality_labels.fits\
-   morphology_catalogue_with_high_quality_labels.csv

Includes: - Object IDs\
- RA, Dec\
- Morphology measurements\
- High-quality labels

------------------------------------------------------------------------

### 3. DINOv2 Embedding Representations

-   File: embeddings.npy\
-   Format: NumPy array\
-   Shape: (365513, D)

Each row corresponds to one galaxy.

------------------------------------------------------------------------

## Data Alignment

Row i in embeddings.npy ↔ row i in catalog ↔ corresponding FITS image.

------------------------------------------------------------------------

## Usage

### Extract images

unzip "\*.zip" -d fits/

### Load embeddings

``` python
import numpy as np
emb = np.load("embeddings.npy")
```

### Load catalog

``` python
import pandas as pd
cat = pd.read_csv("morphology_catalogue_with_high_quality_labels.csv")
```

------------------------------------------------------------------------

## Applications

-   Galaxy morphology classification\
-   Representation learning\
-   Anomaly detection\
-   Similarity retrieval

------------------------------------------------------------------------

## Notes

Derived from Galaxy Zoo Euclid (Q1). Embeddings enable direct ML use
without model inference.