TCM-AQA61-A (Acupuncture) and TCM-AQA61-T (Tuina)
收藏资源简介:
该研究构建了TCM-AQA61-A(针灸)与TCM-AQA61-T(推拿)两个双视角视频数据集,由杜伦大学、首都医科大学等机构联合创建,旨在支持中医康复训练中的动作质量评估。每个数据集均包含61名受试者的同步第一人称与第三人称视频记录,数据规模共计122个样本,并附有专家对分类与连续技能指标的精细标注,涵盖了针刺深度、操作频率等关键评估维度。数据采集过程通过双摄像头同步录制,聚焦于手部精细动作、手物交互及典型自遮挡场景,以真实反映临床操作环境。这些数据集专门用于开发跨视角多模态评估框架,以解决中医手法训练中因复杂手部遮挡和情境依赖导致的评估难题,推动康复技能训练的标准化与智能化发展。
This study developed two dual-perspective video datasets, TCM-AQA61-A (acupuncture) and TCM-AQA61-T (tuina), jointly created by Durham University, Capital Medical University and other institutions. These datasets are designed to support action quality assessment in traditional Chinese medicine (TCM) rehabilitation training. Each dataset contains synchronized first-person and third-person video recordings of 61 subjects, with a total of 122 samples across the two datasets. They are accompanied by fine-grained expert annotations for both categorical and continuous skill indicators, covering key assessment dimensions such as acupuncture depth and manipulation frequency. The data collection was conducted via synchronized dual-camera recording, focusing on fine hand movements, hand-object interactions and typical self-occlusion scenarios to authentically reflect real clinical operating environments. These datasets are specifically developed to facilitate the construction of cross-perspective multimodal assessment frameworks, aiming to address the assessment challenges in TCM manipulation training caused by complex hand occlusions and context dependence, and promote the standardization and intelligent development of rehabilitation skill training.
数据集概述:TCM-AQA61
该数据集是论文 "Towards Cross-View Multimodality Action Quality Assessment for Traditional Chinese Medicine Physical Therapy" 中提出的,用于评估中医理疗动作质量的数据集。
核心内容
- 数据集名称:TCM-AQA61
- 数据规模:包含61名受试者的第一人称和第三人称视角视频。
- 涉及动作:针灸(Acupuncture)和推拿(Chinese Massage/Tui Na)。
- 核心目的:通过跨视角多模态增强框架(CME-AQA)克服单视角骨架数据在中医理疗(如针灸、推拿)中的自遮挡和复杂手物交互问题,提升动作质量评估精度。
数据样例与标签
数据集提供了两类动作的示例视频和标签:
1. 针灸(Acupuncture)
- 视频样本:展示了第三人称视角和第一人称视角的针灸练习场景。
- 评估标签:包括针的持握(NeedleHolding)、进针角度(Angle)、进针深度(Depth)、快速进针(Fast)等10个维度,标签值用0/1表示。例如:
- Video_Side: 1_0_0, Video_Head: 2_0_0, NeedleHolding: 1, Angle: 1, Depth: 1, Fast: 1, InsertionFrequency: 0, InsertionAmplitude: 1, TwistingFrequency: 1, TwistingAmplitude: 1, FastWithdrawal: 1
2. 推拿(Tuina)
- 视频样本:展示了第三人称视角和第一人称视角的推拿练习场景。
- 评估标签:包括动作类型(Action)、手掌空虚(PalmEmpty)、指实(FingerSolid)等7个维度。例如:
- Video_Side: 1_1_0, Video_Head: 2_1_0, Action: 1, PalmEmpty: 0, FingerSolid: 1, Slow: 0, Depth: 0, Frequency: 1
性能表现
所提出的CME-AQA框架在关键评估任务(如 Needle Depth(针刺深度)和 Quick Needle Movements(快速进针动作))上实现了 超过30%的准确率和F1分数提升。
当前状态
- 已发布内容:推理和训练代码以及示例视频片段(包含GIF和MP4文件)。
- 待发布内容:完整的TCM-AQA61数据集将在论文被接收后公开发布。





