Component-Aware Structure-Preserving Style Transfer for Satellite Visual Sim2Real Data Construction
Researchers have developed a novel framework for style transfer to improve the Sim2Real data construction for satellite visual sensing. This method addresses the challenge of acquiring large-scale, accurately annotated real-world satellite images by transferring the appearance of real images to synthetic ones while preserving annotations. The technique uses component-aware, mask-aligned modulation to inject real-domain style codes into synthetic satellite regions, enhancing downstream tasks like pose estimation. AI
IMPACT Enhances the accuracy of satellite pose estimation by improving synthetic-to-real data transfer for training.