MovieNet

Movie understanding. An holistic dataset for movie understanding. In recent years, significant progress has been made in visual understanding. However, how to understand story-based long videos with artistic styles, such as movies, remains challenging. In this paper, we introduce MovieNet—an holistic dataset for movie understanding. MovieNet contains 1,100 movies and includes a large amount of multimodal data, such as trailers, photos, plot descriptions, etc. In addition, MovieNet provides manual annotations from different aspects, including 1.1 million characters with bounding boxes and identities, 42K scene boundaries, 2.5K aligned description sentences, 65K location and action labels, and 92K movie style labels. To our knowledge, MovieNet is the largest dataset with the richest annotations for comprehensive movie understanding. Based on Movie