M-VAD Names (M-VAD Names Dataset)

该数据集包含角色视觉外观的注释，以面部边界框轨迹的形式，以及与角色文本提及的关联（如果可用）。我们通过半自动方法检测和注释每部电影的每个视频剪辑中角色的视觉外观。发布的数据集包含超过 24k 条带注释的视频剪辑，包括 63k 条视觉轨迹和 34k 条文字提及，所有这些都与他们的角色身份相关联。

This dataset provides annotations for characters' visual appearances in the form of facial bounding box trajectories, as well as associations with textual mentions of the characters (if available). We employ a semi-automatic approach to detect and annotate the visual appearances of characters in each video clip of every film. The released dataset contains over 24,000 annotated video clips, including 63,000 visual trajectories and 34,000 textual mentions, all of which are linked to their corresponding character identities.

搜集汇总

背景与挑战

背景概述

M-VAD Names是一个专注于视频字幕生成和角色识别的数据集，包含超过24,000条带注释的视频剪辑，涵盖63,000条视觉轨迹和34,000条文字提及，所有内容均与角色身份关联。该数据集由University of Modena and Reggio Emilia于2018年发布。