TVSeries

“TVSeries 数据集是一个真实的、大规模的动作检测数据集。它由 16 小时的最近六部电视剧的视频组成。定义了 30 个动作类，所有出现的动作都标有开始和结束时间。对于每个动作实例，我们提供了一些元数据（单人、被遮挡、部分动作丢失……），可用于分析方法在特定困难情况下的性能。"

The TVSeries dataset is a real-world, large-scale action detection dataset. It comprises 16 hours of video content sourced from six recent television series. A total of 30 action categories are defined, and all occurring actions are annotated with their precise start and end timestamps. For every action instance, we provide supplementary metadata including single-person involvement, occlusion, partially missing action segments, and other relevant attributes, which can be utilized to analyze the performance of detection methods under specific challenging scenarios.

搜集汇总

背景与挑战

背景概述

TVSeries数据集是一个大规模的动作检测数据集，包含16小时的电视剧视频，标注了30个动作类及其元数据，适用于动作检测和行为预测研究。该数据集由阿姆斯特丹大学和Katholieke Universiteit Leuven于2016年发布。