ITTV - A Dataset of Italian Television for Automatic Genre Classification

ITTV is a publicly available dataset of Italian TV programs introduced in Alessandro Ilic Mezza, Paolo Sani, and Augusto Sarti, "Automatic TV Genre Classification Based on Visually-Conditioned Deep Audio Features," in 2023 31st European Signal Processing Conference (EUSIPCO), 2023. ITTV consists of 2625 manually annotated YouTube videos, totaling over 670 hours. Each clip is assigned one of seven classes: Cartoons Commercials Football Music News Talk Shows Weather Forecast ITTV genre taxonomy is similar to that of the well-known RAI dataset described in Maurizio Montagnuolo and Alberto Messina, "Parallel neural networks for multimodal video genre classification,” Multimedia Tools and Applications, vol. 41, no. 1, pp. 125–159, 2009. The dataset contains genre annotations and metadata in CSV format. Please note that audio data is not provided. We provide the annotations for a balanced training (1575 clips) and validation (525 clips) split, as well as for a disjoint test set containing 525 installments from TV programs not included in the development set. As YouTube continuously updates, some videos may not be available in the future. Although we intend to keep ITTV updated as best as possible, please note that some content may not be available at any given time. Some YouTube videos (especially from the Football class and, to a lesser extent, the Cartoons class) may only be available in some countries due to regional restrictions imposed by the content creator. All videos are known to be accessible from Italy (last accessed on Nov. 25th, 2022.) Please contact Alessandro Ilic Mezza for further questions (e-mail: [e-mail]).

ITTV是一个公开可用的意大利语电视节目数据集，首次由Alessandro Ilic Mezza、Paolo Sani与Augusto Sarti在2023年第31届欧洲信号处理会议（EUSIPCO）上发表的论文《基于视觉条件深度音频特征的自动电视节目分类》中提出。该数据集包含2625条人工标注的YouTube视频，总时长超过670小时。每个视频片段被分配至7个类别之一：卡通片、商业广告、足球赛事、音乐节目、新闻、脱口秀、天气预报。ITTV的分类体系与知名的RAI数据集（RAI dataset）相似，该数据集由Maurizio Montagnuolo与Alberto Messina在2009年发表于《Multimedia Tools and Applications》期刊第41卷第1期的论文《并行神经网络用于多模态视频节目分类》中描述，原文页码为125–159。该数据集以CSV格式提供节目分类标注与元数据。请注意，本数据集不提供音频数据。我们为平衡划分的训练集（1575条片段）、验证集（525条片段）以及独立测试集提供标注；其中测试集包含525条来自开发集未收录电视节目的片段。由于YouTube平台持续更新，部分视频未来可能无法访问。尽管我们将尽可能维护ITTV数据集的可用性，但请注意在任意时间点都可能存在部分内容无法访问的情况。部分YouTube视频（尤其是足球赛事类别，以及程度较轻的卡通片类别）可能因内容创作者设置的地域访问限制，仅在部分国家可访问。已知所有视频均可从意大利境内访问（最后一次可访问时间为2022年11月25日）。如有进一步疑问，请联系Alessandro Ilic Mezza，邮箱：[e-mail]。