TV Series Corpus

<b>Corpus of three TV Series</b> with <b>manual</b> and <b>automatic</b> annotations:<br><br>- <i>Breaking Bad</i>: S01--S03 (file 'bb.json')<br>- <i>Game of Thrones</i>: S01--05 (file 'got.json')<br>- <i>House of Cards</i>: S01--S02 (file 'hoc.json')<br><br>All three files are in .json format and contain TV Series <b>projects</b>.<br><br>A <b>project</b> is specified by a <b>name</b> and is related to a TV Series defined by its <b>name</b>,<br><br>A TV Series contains <b>seasons</b>, defined by their <b>number</b>.<br><br>Every season is made of <b>episodes</b>, defined by their <b>name</b> and <b>number</b>, along wih the <b>name</b> of the associated video file, its <b>frame rate</b> and size (<b>width</b> x <b>height</b>).<br><br>Each episode contains both <b>scenes</b> an <b>subtitles</b>.<br><br>Scenes are made of <b>shots.<br><br></b>A shot is defined by<b>:<br><br></b>-<b> Cam</b>era position labels, both manual and automatic.<br>- Starting and <b>end</b>ing <b>pos</b>itions.<br>- <b>Music</b>ality rate measured every second and ranging from 0 to 1.<br>- Coordinates of rectangular boxes surrounding possible <b>faces</b>. Face detection is performed on a sample of 5 frames uniformly distributed over the shot. The <b>pos</b>ition of each of these frames is mentioned.<br><br>The subtitles are defined by their:<br><br>- <b>Text</b>ual content.<br>- <b>Sp</b>ea<b>k</b>ing character.<br>- Possible<b> inter</b>locutor(s): manually annotated; automatically annotated from the subtitle sequence, automatically annotated from the characters co-appearing in the current scene.<br><br>All timestamps are expressed in milliseconds and are valid for the video files extracted from the commercial DVDs, with recaps included at the beginning of the <i>House of Cards</i> episodes.<br><br>The .pdf file details the <b>source of annotation</b> (manual/automatic) for each of the tasks considered. <b> <br> </b><br>

<b>三部电视剧语料库（Corpus）</b>，附带<b>人工标注</b>与<b>自动标注</b>结果：<br><br>- <i>《绝命毒师（Breaking Bad）》</i>：第1至3季（文件bb.json）<br>- <i>《权力的游戏（Game of Thrones）》</i>：第1至5季（文件got.json）<br>- <i>《纸牌屋（House of Cards）》</i>：第1至2季（文件hoc.json）<br><br>所有三个文件均采用JSON格式，其中包含电视剧<b>项目（project）</b>。<br><br><b>项目（project）</b>以<b>名称</b>作为唯一标识，并与以<b>名称</b>定义的单部电视剧相关联，<br>单部电视剧包含按<b>编号</b>定义的<b>季（season）</b>。<br><br>每一季由按<b>名称</b>与<b>编号</b>定义的<b>剧集（episode）</b>组成，同时附带对应视频文件的<b>名称</b>、<b>帧率（frame rate）</b>以及分辨率尺寸（<b>宽（width）</b> × <b>高（height）</b>）。<br><br>每一集同时包含<b>场景（scene）</b>与<b>字幕（subtitle）</b>。<br><br>场景由<b>镜头（shot）</b>构成。<br><br><b>镜头（shot）</b>的定义要素如下：<br><br>- <b>摄像机（Camera）</b>位置标签，涵盖人工标注与自动标注结果。<br>- <b>起始与终止位置</b>。<br>- <b>音乐性评分</b>：每秒测算一次，取值范围为0至1。<br>- 环绕可能存在<b>人脸（face）</b>的矩形框坐标。人脸检测在镜头内均匀分布的5帧采样上完成，各采样帧的<b>位置</b>均已标注。<br><br><b>字幕（subtitle）</b>的定义要素如下：<br><br>- <b>文本内容</b>。<br>- <b>发言角色</b>。<br>- 潜在<b>对话者（interlocutor）</b>：包含人工标注结果、基于字幕序列自动生成的标注结果，以及基于当前场景中共同出现的角色自动生成的标注结果。<br><br>所有<b>时间戳</b>均以毫秒为单位，适用于从商业DVD提取的视频文件——<i>《纸牌屋（House of Cards）》</i>的剧集开头包含剧集回顾片段，该片段的时间戳同样有效。<br><br>附带的<b>PDF</b>文件详细说明了各标注任务的<b>标注来源</b>（人工/自动）。<b><br></b><br>