FineGym 动作识别数据集

FineGym 是一个以体操视频为基础的动作识别数据集。与现有的同类型数据集相比，FineGym 数据集更加丰富、多样，而且品质更高。该数据集具有三级语义层级结构，给动作级和子动作级都提供了时间标注。例如一个 “平衡木” 事件，可以被标注为一系列来自五个集合的基本子动作，而且这些集合中的子动作都将进一步被类标签标注。

FineGym is an action recognition dataset built on gymnastics videos. Compared with existing comparable datasets, FineGym is more comprehensive, diverse and higher in quality. This dataset features a three-level semantic hierarchy, providing temporal annotations at both the action and sub-action levels. For example, a "balance beam" event can be annotated as a series of basic sub-actions from five sets, and each sub-action within these sets will be further labeled with class tags.

搜集汇总

背景与挑战

背景概述

FineGym是一个专注于体操视频的动作识别数据集，其特点是具有三级语义层级结构，为动作和子动作提供了时间标注。该数据集在丰富性、多样性和质量方面优于同类现有数据集，例如将“平衡木”事件分解为基本子动作并进行分类标注。