zyyy12138/motion-smd-data

Motion-SMD数据集是一个用于人类运动理解的结构化运动描述和预处理运动语言数据的数据集。它包含四个子目录，分别提供预生成的结构化运动描述JSON文件、BABEL-QA和HuMMan-QA的关节数据和问答数据、以及HumanML3D的22关节序列和文本数据。这些数据支持运动描述生成、运动问答等任务，旨在通过结构化文本描述实现无需编码器的人类运动理解。数据集的主要贡献包括规则生成的结构化运动描述和10选项的问答数据。数据集的大小在10K到100K之间，使用混合许可证，具体许可证信息见各子目录的README文件。

The Motion-SMD dataset is a collection of structured motion descriptions and preprocessed motion-language data for human motion understanding. It includes four subdirectories containing pre-generated Structured Motion Description JSONs, joint data and QA data from BABEL-QA and HuMMan-QA, and 22-joint sequences and texts from HumanML3D. The dataset supports tasks such as motion captioning and motion QA, aiming to achieve encoder-free human motion understanding via structured text descriptions. Key contributions include rule-based structured motion descriptions and 10-option QA data. The dataset size ranges between 10K and 100K, and it is licensed under a mixed license, with details provided in each subdirectorys README file.