SARLO-80: Worldwide Slant SAR Language Optic Dataset - 80cm

Description

This dataset provides 119566 paired Umbra SAR, optical images in slant-range geometry and text description.

• SAR data come from Umbra SICD products, resampled to a fixed 80 cm × 80 cm resolution (azimuth and range) in slant-range geometry.
• Optical images are high-resolution optical data reprojected into the slant-range geometry of the corresponding SAR acquisition.
• Each pair also includes three automatically generated captions from the optical image using CogVLM2, then cleaned and post-processed with the Qwen LLM.

Format

• Resolution: 80 cm (azimuth × range), slant-range geometry
• Modality pairing: Umbra SAR × optical (slant-range reprojected)
• Text: 3 captions per pair (SHORT, MID, LONG)

Sample contents (per WebDataset shard)

Each sample contains:

The meta.json contains the information of optic.png, city, city_id_satellite_pass, SAR geometry: Nb_pixel_a, Nb_pixel_r, spacing_eff_az, spacing_eff_rg, operation_sampling, bbox_ecf, bbox_llh and caption: SHORT – short, MID – medium-length and LONG – descriptive caption.

More examples

Usage (WebDataset / Python)

Example to stream the dataset with webdataset:

import os, io
import numpy as np
from PIL import Image
from huggingface_hub import HfApi, hf_hub_download
import webdataset as wds

token = os.environ.get("HF_TOKEN") #put your token_hf

repo_id = "ONERA/SARLO-80"

chunk_id = 0
shard_name = "shard-00027.tar"  # exact name
filename = f"train/chunk_{chunk_id:03d}/{shard_name}"


local_tar = hf_hub_download(
            repo_id=repo_id,
            repo_type="dataset",
            filename=filename,
            token=token,
        )

ds = wds.WebDataset(local_tar, shardshuffle=False)
sample = next(iter(ds))

img_optic = Image.open(io.BytesIO(sample["optic.png"]))
print("  optic.png size (W,H):", img_optic.size)

img_sar = Image.open(io.BytesIO(sample["sar.png"]))
print("  sar.png   size (W,H):", img_sar.size)

sar_complex_slc = np.load(io.BytesIO(sample["sar.npy"]), allow_pickle=False)
print("  sar.npy dtype/shape :", sar_complex_slc.dtype, sar_complex_slc.shape)

meta_bytes = sample["meta.json"]          
meta = json.loads(meta_bytes.decode("utf-8"))
print(f'Metadata: {meta}')

Acknowledgments

This work was carried out as part of the PhD of Solène Debuysère at DEMR-ONERA - Université de Paris Saclay, under the supervision of Nicolas Trouvé, Nathan Letheule, and Elise Colin. We gratefully acknowledge ONERA, and especially DEMR-SEM Team for providing computational and research resources, Umbra for the SAR data collections and open-access initiatives enabling research use, and Hugging Face, in particular Georgia Channing, for her help on this project.

Contacts: [email protected] or [email protected]

License

SARLO-80 is released under the Creative Commons Attribution‑ShareAlike 4.0 (CC‑BY‑SA‑4.0) license.

Citation information will be made available upon soon paper publication.