FSAR-Cap: A Fine-Grained Two-Stage Annotated Dataset for SAR Image Captioning

Synthetic Aperture Radar (SAR) image captioning facilitates scene-level semantic understanding and plays a critical role in military intelligence and urban planning; however, its progress is constrained by the scarcity of high-quality annotated datasets. To address this limitation, we present FSAR-Cap, a large-scale SAR captioning dataset with 14,480 images and 72,400 image–text pairs. FSAR-Cap is built on the FAIR-CSAR detection dataset and constructed through a two-stage annotation strategy that combines hierarchical template-based representation, manual verification, supplementation and prompt standardization. Compared with existing resources, FSAR-Cap provides richer fine-grained annotations. Benchmarking with multiple encoder–decoder architectures and Visual-language models(VLMs) verifies its effectiveness, establishing a foundation for future research in SAR captioning and image interpretation. The dataset will be publicly available at: https://github.com/hitjiao/FSAR-Cap.