MultiSports

多运动是一个时空行为检测数据集。数据集具有以下特征:
多人场景: 在同一场景中，不同的人执行不同的细粒度动作，减少了背景提供的信息。行动边界是明确的。相对细粒度的细粒度动作类别。
通过选择4个体育类别，数据集收集3,200视频片段以构建多体育v1.0数据集。数据集还用902k个边界框注释37,701操作实例。数据集具有高度多样性、密集注释和高质量的重要特性。

MultiSports is a spatio-temporal behavior detection dataset. The dataset has the following characteristics:
Multi-person scenarios: In the same scene, different individuals perform distinct fine-grained actions, which reduces the information provided by the background. Action boundaries are clearly defined, and there are relatively fine-grained action categories.
By selecting 4 sports categories, 3,200 video clips were collected to construct the MultiSports v1.0 dataset. The dataset also annotates 37,701 action instances with 902k bounding boxes. It features high diversity, dense annotations, and high quality.

搜集汇总

背景与挑战

背景概述

MultiSports是一个细粒度时空行为检测数据集，包含4个体育类别的3,200个视频片段和37,701个动作实例的高质量注释，适用于多人场景下的行为分析研究。