YouCook

该数据集是从 88 个开源 YouTube 烹饪视频中准备的。 YouCook 数据集包含人们烹饪各种食谱的视频。视频是从YouTube上下载的，都是第三人称视角；它们代表了比现有的烹饪和厨房数据集更具挑战性的视觉问题（背景厨房/场景对于许多人来说是不同的，并且大多数视频都有动态的相机变化）。此外，还为训练数据（以及许多预先计算的低级特征）提供逐帧对象和动作注释。最后，每个视频都有许多人工提供的自然语言描述（平均而言，每个视频有八种不同的描述）。该数据集的创建是为了作为使用自然语言描述描述复杂现实世界视频的基准。

This dataset is compiled from 88 open-source YouTube cooking videos. The YouCook Dataset contains videos of people preparing a wide range of recipes. All videos are downloaded from YouTube and captured from a third-person perspective. They present more challenging visual scenarios than existing cooking and kitchen datasets, as the background kitchen environments or scenes differ across videos, and most feature dynamic camera movements. Additionally, frame-by-frame object and action annotations are provided for the training split, alongside numerous pre-computed low-level features. Finally, each video is paired with multiple human-authored natural language descriptions—on average, eight distinct descriptions per video. This dataset was developed to act as a benchmark for describing complex real-world videos using natural language.

搜集汇总

背景与挑战

背景概述

YouCook数据集基于88个YouTube烹饪视频构建，提供第三人称视角视频、逐帧对象与动作注释，以及人工撰写的自然语言描述。该数据集旨在作为使用自然语言描述复杂现实世界视频的基准。