Fruit-Vegtable Grasping Dataset

FruitVegGrasp is the first open-access dataset for semantic grasp-region modeling in agriculture, comprising 54,092 images of 27 crop types (17 fruits and 10 vegetables) across real, synthetic, and style-transferred domains. Each instance is annotated with six pixel-wise grasp regions (e.g., top_left, side_right) following a unified JSON schema, enabling fine-grained imitation of human harvesting decisions. Designed to address the limitations of industrial-object datasets, FruitVegGrasp supports research on semantic segmentation, domain adaptation, and grasp planning, offering a robust benchmark for developing intelligent, data-driven harvesting robots.