Control Copy-Paste: Controllable Diffusion-Based Augmentation Method For Remote Sensing Few-Shot Object Detection

Few-shot object detection in optical remote sensing imagery faces challenges due to limited labeled training data. This paper proposes Control Copy-Paste, a controllable diffusion-based data augmentation method that leverages diffusion models to generate diverse and realistic training samples for remote sensing few-shot object detection. The method addresses overfitting issues caused by scarce training data by synthesizing high-quality augmented images with controllable spatial layouts, improving detector generalization in low-data regimes.