Dataset Card: SIVED (SAR Image Vehicle Detection)

Important Notice

This dataset was NOT created by the uploader. It was originally curated and published by the researchers listed below. It is shared here on Kaggle and Hugging Face solely for accessibility and ease of use in machine learning pipelines. All credit belongs to the original authors.

Authors

Xin Lin, Bo Zhang, Fan Wu, Chao Wang, Yali Yang, and Huiqin Chen

CAESAR-Radi Research Group

Original Sources

Source	Link
GitHub Repository	https://github.com/CAESAR-Radi/SIVED
Paper (Remote Sensing, MDPI)	https://www.mdpi.com/2072-4292/15/11/2825
Google Drive (original download)	https://drive.google.com/drive/folders/1ntqNSMQqMwu8RE4uK2z632cMQEi4yid6

Dataset Description

SIVED is a SAR (Synthetic Aperture Radar) image dataset designed for vehicle detection using rotatable bounding box annotations. It aggregates high-resolution SAR imagery from three distinct radar sources across multiple frequency bands.

Property	Value
Total Images	1,044 (512 x 512 px, grayscale)
Total Annotations	12,013 oriented bounding boxes
Classes	1 (Vehicle)
Annotation Format	DOTA (8-point rotated bounding box)
Splits	Train: 837 / Valid: 104 / Test: 103

Radar Sources

Source	Organization	Band	Polarization	Resolution
FARAD	Sandia National Laboratory	Ka/X	VV/HH	0.1m x 0.1m
MiniSAR	Sandia National Laboratory	Ku	-	0.1m x 0.1m
MSTAR	U.S. Air Force	X	HH	0.3m x 0.3m

Annotation Format

Each annotation provides eight coordinates defining the four corners of an oriented bounding box, a class label, and a difficulty flag:

x1 y1 x2 y2 x3 y3 x4 y4 class_name difficulty

• difficulty=0: Easy (clearly visible vehicles)
• difficulty=1: Hard (partially occluded, low contrast, or very small)

License

The original SIVED repository on GitHub does not specify an explicit license file. The accompanying paper was published in Remote Sensing (MDPI), an open-access journal that publishes under the Creative Commons Attribution (CC BY 4.0) license, which permits sharing and adaptation with appropriate credit.

Users of this dataset should cite the original paper and comply with any terms set by the original authors. If there is any concern regarding redistribution rights, please refer to the original repository or contact the authors directly.

Citation (Required)

If you use this dataset in any capacity, please cite the original work:

@Article{rs15112825,
  author  = {Lin, Xin and Zhang, Bo and Wu, Fan and Wang, Chao and Yang, Yali and Chen, Huiqin},
  title   = {SIVED: A SAR Image Dataset for Vehicle Detection Based on Rotatable Bounding Box},
  journal = {Remote Sensing},
  volume  = {15},
  number  = {11},
  pages   = {2825},
  year    = {2023},
  doi     = {10.3390/rs15112825},
  url     = {https://www.mdpi.com/2072-4292/15/11/2825}
}

Related Work

• Scientific Reports (Nature, 2025): https://www.nature.com/articles/s41598-025-28755-3