KIT Motion-Language

该数据集名为KIT Motion-Language数据集，包含运动捕捉的动作和参考描述。具体来说，它由3911个运动序列组成，每个序列由一系列50维向量有序排列表示。此外，该数据集还包含了一个子集，其中包含2721个动作和5538个用于训练、验证和评估的描述句子。每个动作都配有多个描述，而且运动数据的值已被归一化至[-0.9, 0.9]的范围。在规模上，数据集包含3911个运动序列和6353个描述句子。其任务是进行描述与动作之间的双向翻译。

This dataset is named the KIT Motion-Language Dataset, which contains motion capture action data and their reference descriptions. Specifically, it consists of 3911 motion sequences, each represented by an ordered sequence of 50-dimensional vectors. Additionally, the dataset includes a subset that contains 2721 actions and 5538 descriptive sentences for training, validation, and evaluation. Each action is paired with multiple descriptive sentences, and all motion data values have been normalized to the range [-0.9, 0.9]. In terms of scale, the dataset totals 3911 motion sequences and 6353 descriptive sentences. The task of this dataset is to perform bidirectional translation between textual descriptions and motion actions.

搜集汇总

背景与挑战

背景概述

KIT Motion-Language数据集是一个用于运动标注的工具项目，已存档并迁移到新地址。它提供在线访问链接和关联论文，支持Python环境下的安装和配置，包括依赖管理、数据库设置和运动数据导入，需要KIT Whole-Body Human Motion Database账户，最后更新于2017年。