Remote Sensing Target Forgery Dataset ( RSTFD)

To address the limitations of existing remote sensing forgery datasets (scarcity, single forgery type, lack of pixel-level annotations), we propose RSTFD\u2014a comprehensive target-oriented benchmark for forgery detection. It includes 1,626 512\u00d7512 images (3:1 train-test split) covering three high-value targets (civilian\/military aircraft, ships) and three forgery types (splicing, copy-move, AIGC removal). Built from three open-source datasets, RSTFD adopts COCO-style polygon annotations (auto-generated via Segment Anything + manual refinement) and strict quality filtering. Statistically balanced in forgery types, it covers multi-scale forgery regions (0.15%\u201315.19% pixel proportion). Publicly available with generation scripts, RSTFD fills the gap of target-oriented multi-type remote sensing forgery benchmarks, supporting advanced detection model development.