IO-AI Office PicknPlace

IO-AI Office PicknPlace 数据集是面向办公及零售场景的机器人自动化分拣与摆放任务的专用数据集，支持基于语义标签（如品牌名称、条形码）的位姿感知与抓取策略训练。其核心包含真实环境的多视角RGB-D图像（4.6万张）和Unity生成的虚拟场景（12.9万张），覆盖20类常见商品，并通过全卷积网络实现语义与几何联合分析，优化复杂遮挡场景下的双机械臂协同操作。该数据集通过真实与虚拟数据融合，为仓储物流、零售分拣等场景的算法开发提供标准化测试平台。

The IO-AI Office PicknPlace Dataset is a specialized dataset for robotic automated sorting and placement tasks in office and retail scenarios. It supports pose perception and grasp strategy training based on semantic labels (e.g., brand names, barcodes). Its core consists of 46,000 multi-view RGB-D images captured in real environments and 129,000 virtual scenes generated by Unity, covering 20 categories of common commodities. It implements joint semantic and geometric analysis via fully convolutional networks (FCNs) to optimize the collaborative operation of dual robotic arms in complex occlusion scenarios. By fusing real and virtual data, this dataset provides a standardized testing platform for algorithm development in scenarios such as warehouse logistics and retail sorting.