ryushinn/Taichi-HD-BBox-Keypoint

该数据集包含从`ryushinn/Taichi-HD`数据集估计的人物边界框和COCO-17身体关键点，是一个仅包含注释的辅助数据集，不复制源图像。数据集的结构包括训练集和测试集的行数及检测到的人物行数，每行包含四个列：边界框坐标、边界框置信度、关键点坐标和关键点置信度。关键点遵循COCO身体的17个关键点顺序，并提供了用于可视化的骨架边。数据集的注释是通过OpenMMLab模型生成的，包括RTMDet-M COCO人物检测器和RTMPose-M COCO身体姿态估计器。数据集的使用方法包括加载源图像和注释数据集，并通过行索引进行配对。

This dataset contains estimated person bounding boxes and COCO-17 body keypoints for `ryushinn/Taichi-HD`. It is an annotations-only sidecar dataset that does not duplicate the source images. The dataset structure includes the number of rows and detected person rows for the train and test splits, with each row containing four columns: bounding box coordinates, bounding box confidence score, keypoint coordinates, and keypoint confidence scores. The 17 keypoints follow the COCO body convention, and skeleton edges for visualization are provided. The annotations were generated using OpenMMLab models, specifically the RTMDet-M COCO person detector and RTMPose-M COCO body pose estimator. Usage examples demonstrate how to load and pair the source images with the annotation dataset.