SAR-1M: 1 Million SAR Images with 750,000 Aligned Optical Pairs

SAR-1M contains over 1,000,000 synthetic aperture radar (SAR) images for remote sensing representation learning, released by Wenquandan777. Approximately 75% of the SAR samples are paired with geographically aligned optical images to support multimodal analysis.

Use Cases

• Self-supervised representation learning using the 1,000,000+ SAR images
• Cross-modal translation between the SAR and OPT image directories
• Multimodal feature extraction using geographically aligned image pairs

Strengths

• Scale of over 1,000,000 SAR images
• Geographically aligned optical pairs for 75% of the data
• Standardized PNG format for both SAR and OPT modalities

Limitations

• Missing optical pairs for approximately 250,000 SAR images
• Absence of semantic class labels or bounding box annotations

Provenance

Source

Wenquandan777 (ArXiv:2512.16635)

Geography

United States