"DiffSAR-FS benchmark"

"Cross-modality few-shot SAR detection aims to leverage abundant optical imagery to improve target detection in few-shot SAR detection. Despite its practical significance, there is a lack of standardized benchmarks to systematically evaluate cross-modal transfer under few-shot settings. We construct a benchmark covering four SAR detection tasks, including ship, vehicle, bridge, and aircraft detection. Each benchmark integrates a SAR dataset for few-shot learning with a large-scale optical dataset as auxiliary supervision, while maintaining consistent category alignment across modalities. Specifically, we pair SSDD with HRRSD for ship detection, SIVED with HRRSD for vehicle detection, OGSOD for bridge detection with aligned optical\u2013SAR pairs, and SAR-AIRcraft-1.0 with DIOR for aircraft detection. Few-shot samples are drawn exclusively from SAR training sets, and evaluation is conducted on corresponding SAR test sets to reflect realistic deployment conditions. The constructed benchmarks exhibit diverse scene distributions, scale variations, and modality gaps, providing a unified testbed for studying cross-modality knowledge transfer in few-shot SAR detection."