---
license: cc-by-sa-4.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: astropt_15m_hsc
    list: float32
    length: 384
  - name: astropt_95m_hsc
    list: float32
    length: 768
  - name: astropt_850m_hsc
    list: float32
    length: 2048
  - name: convnext_nano_hsc
    list: float32
    length: 640
  - name: convnext_tiny_hsc
    list: float32
    length: 768
  - name: convnext_base_hsc
    list: float32
    length: 1024
  - name: convnext_large_hsc
    list: float32
    length: 1536
  - name: dino_small_hsc
    list: float32
    length: 384
  - name: dino_base_hsc
    list: float32
    length: 768
  - name: dino_large_hsc
    list: float32
    length: 1024
  - name: dino_giant_hsc
    list: float32
    length: 1536
  - name: ijepa_huge_hsc
    list: float32
    length: 1280
  - name: ijepa_giant_hsc
    list: float32
    length: 1408
  - name: vit_base_hsc
    list: float32
    length: 768
  - name: vit_large_hsc
    list: float32
    length: 1024
  - name: vit_huge_hsc
    list: float32
    length: 1280
  - name: specformer_base_desi
    list: float64
    length: 768
  splits:
  - name: train
    num_bytes: 1487887360
    num_examples: 20465
  download_size: 1676612642
  dataset_size: 1487887360
---

# DESI–HSC Embeddings

This dataset contains precomputed embeddings for cross-survey sources (DESI ↔ HSC).  
Each row includes one object ID and multiple embedding vectors from different backbone models (e.g. DINO, ViT, AstroPT).

## Load in Python

```python
from datasets import load_dataset
import numpy as np

ds = load_dataset("UniverseTBD/desi_hsc_embeddings", split="train")

print("Columns:", ds.column_names)   

# Choose one HSC image embedding and the DESI spectral embedding:
col_a = "specformer_base_desi"   # DESI spectra (SpecFormer)
col_b = "dino_large_hsc"         # HSC images (DINOv2 Large)
assert col_a in ds.column_names and col_b in ds.column_names


```