# S1 Single-Ship: Sentinel-1 SAR Ship Detection Downstream Task (SSL4EO-S12 / NeuCo-Bench)

## Facts

- Provider: Embed2Scale consortium (DLR, IBM Research Europe and partners)
- License: CC-BY-4.0 (https://huggingface.co/datasets/embed2scale/S1-single-ship-downstream)
- Size: 6,036 files
- Format: TIFF, CSV (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/embed2scale/S1-single-ship-downstream
- Access: public, readable without an account
- Category: Earth Observation & Geospatial
- Source: https://arxiv.org/abs/2503.00168
- Page: https://scigantic.com/archive/bdd4d383-6ba8-5bd0-86eb-d6bad2620433
- Keywords: Earth Observation & Geospatial, satellite imagery, TIFF, CSV, Hugging Face

## About

A maritime downstream task for Earth-observation foundation models and neural compression: 2,010 Sentinel-1 GRD image cubes (VV and VH polarisation, GeoTIFF), each containing exactly one verified ship. Ship positions come from public AIS vessel-traffic broadcasts (NOAA Marine Cadastre and US Coast Guard) around 13 US ports, matched to the Sentinel-1 acquisition time. Most chips are over open water, with a subset over rivers or near land. A metadata CSV gives the ship bounding box (WKT), pixel and latitude/longitude position, S1 timestamp and AIS attributes (vessel type code, name, IMO, MMSI, length, width, draft, cargo). For compatibility with the SSL4EO-S12 v1.1 layout and the NeuCo-Bench framework the release also ships 264x264 Sentinel-2 L1C and L2A cubes chosen to be fully cloud covered, so any model using both modalities must keep the ship signal from SAR; S2 is stored as zipped Zarr. Folders: data/ (S1 and S2 cubes), labels/ (NeuCo-Bench task files), ship_labels_metadata.csv and a dataloader.py. About 14 GB in total, of which S1 is 2.3 GB. Licensed CC BY 4.0 by the authors; the underlying NOAA AIS data are US public domain. Cite the Embed2Scale SSL4EO-S12 v1.1 paper (Blumenstiel et al., 2025).

## Load the data

```python
import pandas as pd
df = pd.read_csv("hf://datasets/embed2scale/S1-single-ship-downstream/labels/vessel_x__regr.csv")
df.head()
```

## Data structure

File types in sampled files: TIFF, CSV

Example file: `labels/vessel_x__regr.csv`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/bdd4d383-6ba8-5bd0-86eb-d6bad2620433
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/bdd4d383-6ba8-5bd0-86eb-d6bad2620433
- Over MCP: `get_schema_card` with id `bdd4d383-6ba8-5bd0-86eb-d6bad2620433` at https://api.scigantic.com/mcp

## Related topics

- [Earth Observation & Geospatial datasets](https://scigantic.com/topics/earth-observation-and-geospatial)
- [CSV datasets](https://scigantic.com/topics/csv)
- [Satellite imagery datasets](https://scigantic.com/topics/satellite-imagery)
- [TIFF datasets](https://scigantic.com/topics/tiff)
