YouCook2 is the largest task-oriented instructional video dataset in the visual community. The dataset consists of 2,000 YouTube videos of 89 recipes, all of which are uncut, long videos. The total video length is 176 hours, and the average length of a single video is 5.26 minutes. Each video is shot within 10 minutes and is shot by an individual at home with a non-fixed camera.

On average, there are 22 videos for each recipe. The procedural steps of each video are annotated with time boundaries and described in English sentences. The videos are downloaded from YouTube and are all shot from a third-person perspective. The YouCook2 dataset contains a rich variety of recipe types and a variety of cooking styles from around the world.