用于6D位姿估计的高反射无纹理工件数据集（HW6D）

这是一个自制的数据，用于评估名为 Swin-AFF 的 6D 位姿估计网络，用于高反射和无纹理工件。在这个数据集中，法兰和连接器是具有反射表面和对称形状的金属工件，而框架和夹具是没有纹理的3D打印塑料工件，大小和形状各不相同。Intel RealSense D435i 摄像头用于收集 8 个视频。对目标工件在每一帧中的实际姿态进行标记，从而为每个工件生成大约 6,000 个标记图像。根据LineMOD对数据集进行划分，并为每个对象额外添加20,000个合成数据以扩展数据集。

This is a self-made dataset for evaluating the Swin-AFF 6D pose estimation network, designed for highly reflective and texture-less industrial workpieces. In this dataset, flanges and connectors are metallic workpieces with reflective surfaces and symmetric shapes, while frames and fixtures are texture-less 3D-printed plastic workpieces with varying sizes and shapes. Intel RealSense D435i cameras were used to collect 8 video sequences. The ground-truth poses of the target workpieces in each frame were annotated, generating approximately 6,000 annotated images per workpiece. The dataset was split following the LineMOD protocol, and an additional 20,000 synthetic samples were added for each object to expand the dataset.