SSDD: SAR Ship Detection Dataset (Object Detection)

Unofficial redistribution of SSDD (SAR Ship Detection Dataset), reformatted into a standardized YOLO-compatible directory layout with a seeded validation split.

Dataset Description

• Homepage: https://github.com/TianwenZhang0825/Official-SSDD
• Repository: https://github.com/TianwenZhang0825/Official-SSDD
• Paper: https://doi.org/10.3390/rs13183690

Disclaimer

This repository is not an official release of SSDD.

SSDD was created by Tianwen Zhang, Xiaoling Zhang, Jianwei Li, and co-authors, who retain all copyright. This repository does not claim ownership of any images, annotations, or metadata.

This repository exists for two purposes:

1. To reorganize the official BBox_SSDD/coco_style release into a standardized YOLO/Ultralytics-compatible directory structure.
2. To provide a defined train/val/test split, since the official release ships train/test only (see Changes from the Official Release below).

Direct provenance. This redistribution is sourced directly from the official GitHub repository (TianwenZhang0825/Official-SSDD), which links to the actual data on Google Drive / Baidu Cloud. This repository converts the already-COCO-formatted BBox_SSDD/coco_style variant to YOLO format and applies a seeded validation split.

---

Dataset Overview

SSDD (SAR Ship Detection Dataset) is a benchmark for ship detection in Synthetic Aperture Radar imagery: 1,160 images with 2,587 ship instances, composited from RadarSat-2, TerraSAR-X, and Sentinel-1 at resolutions from 1m to 15m, across multiple polarizations (HH/VV/VH/HV) and both inshore and offshore scenes. It is used to benchmark SAR ship detection, where speckle noise and side-lobe artifacts make optical-trained detectors unreliable -- the same problem HRSID targets, from a different sensor mix.

This repository preserves every image and box while re-encoding the labels for YOLO compatibility and adding a reproducible validation split (see Changes from the Official Release below).

---

Changes from the Official Release

• Validation split carved out of train. The official release provides only train (928 images) and test (232 images). This repository holds out a seeded random 15% of train as valid (139 images), leaving 789 for train; test is kept as-is, unchanged.
• Format converted. The official release already ships COCO-format annotations (annotations/{train,test}.json, single class ship already at id 0); this repository additionally provides a YOLO-format export (normalized class x_center y_center width height .txt files), alongside a re-split canonical COCO layout.
• test_inshore / test_offshore not separately included. These are filtered subsets of test provided by the official release for scene-specific analysis, not additional data -- everything in both subsets is already covered by test.
• No image pixel content was modified. No boxes were added or removed.

---

Dataset Structure

<repo>/
├── README.md
├── ssdd_banner.jpg
└── data/
    ├── data.yaml
    ├── images/
    │   ├── train/   (789 *.jpg)
    │   ├── valid/   (139 *.jpg)
    │   └── test/    (232 *.jpg)
    └── labels/
        ├── train/   (789 *.txt)
        ├── valid/
        └── test/

• data/images/<split>/ contains the SAR ship-detection images for each split.
• data/labels/<split>/ contains one YOLO-format .txt annotation file per image (class x_center y_center width height, normalized; empty for images with no annotated ship).
• data/data.yaml is the Ultralytics dataset configuration file.
• Splits: train 789 images / 1,756 boxes · valid 139 images / 285 boxes · test 232 images / 546 boxes (1,160 images / 2,587 boxes total).

Classes (1)

ship -- single class.

Full split summary, class-distribution chart, and box geometry: see the dataset statistics report.

---

Dataset Sources

Original Paper

SAR Ship Detection Dataset (SSDD): Official Release and Comprehensive Data Analysis

Tianwen Zhang, Xiaoling Zhang, Jianwei Li, Xiaowo Xu, Baoyou Wang, Xu Zhan, Yanqin Xu, Xu Ke, Tianjiao Zeng, Hao Su, and others

Remote Sensing, 13(18):3690, 2021.

Official Resources

• GitHub Repository: https://github.com/TianwenZhang0825/Official-SSDD

---

Attribution

All credit for the dataset belongs entirely to the original SSDD authors: Tianwen Zhang, Xiaoling Zhang, Jianwei Li, and their co-authors.

This repository only reformats the official release into a YOLO layout and adds a seeded validation split.

If you use this dataset in your research, please cite the original publication below.

---

License

The official GitHub repository carries an explicit Apache License 2.0 (LICENSE file, confirmed via GitHub's own license detection -- not just a badge claim). This is a real, unambiguous grant, unlike HRSID's software-only GPL-3.0 situation.

A caveat worth knowing. SSDD's imagery is composited from RadarSat-2, TerraSAR-X, and Sentinel-1. TerraSAR-X / TanDEM-X data is distributed by DLR under a scientific-use license that does not, in general, permit free redistribution of derived image products -- independent of what the repackager's own repo license grants. It is not possible to determine from public information which images derive from which sensor, or whether the SSDD authors obtained separate redistribution rights for the TerraSAR-X-derived portion specifically. This is the same second-order sensor-rights situation documented on HRSID's card.

Given the repo's own license is explicit (unlike HRSID's), this repository is tagged apache-2.0 rather than unknown. If you plan to use this dataset for anything beyond personal research experimentation, we recommend being aware of the TerraSAR-X provenance question above.

Accordingly:

• Attribution to the original creators is required.
• Commercial and non-commercial use are both permitted per the stated Apache-2.0 terms.
• No share-alike obligation.

---

Citation

If you use this dataset, please cite:

@article{zhang2021sar,
  title={SAR Ship Detection Dataset (SSDD): Official Release and Comprehensive Data Analysis},
  author={Zhang, Tianwen and Zhang, Xiaoling and Li, Jianwei and Xu, Xiaowo and Wang, Baoyou and Zhan, Xu and Xu, Yanqin and Ke, Xu and Zeng, Tianjiao and Su, Hao and others},
  journal={Remote Sensing},
  volume={13},
  number={18},
  pages={3690},
  year={2021}
}

---

Acknowledgements

We sincerely thank Tianwen Zhang, Xiaoling Zhang, Jianwei Li, and their co-authors for creating and publicly releasing this valuable SAR ship-detection benchmark.