HLVU: A New Challenge to Test Deep Understanding of Movies the Way Humans do

Abstract

In this paper we propose a new evaluation challenge and direction in the area of High-level Video Understanding. The challenge we are proposing is designed to test automatic video analysis and understanding, and how accurately systems can comprehend a movie in terms of actors, entities, events and their relationship to each other. A pilot High-Level Video Understanding (HLVU) dataset of open source movies were collected for human assessors to build a knowledge graph representing each of them. A set of queries will be derived from the knowledge graph to test systems on retrieving relationships among actors, as well as reasoning and retrieving non-visual concepts. The objective is to benchmark if a computer system can "understand" non-explicit but obvious relationships the same way humans do when they watch the same movies. This is long-standing problem that is being addressed in the text domain and this project moves similar research to the video domain. Work of this nature is foundational to future video analytics and video understanding technologies. This work can be of interest to streaming services and broadcasters hoping to provide more intuitive ways for their customers to interact with and consume video content. CCS CONCEPTS• Information systems → Retrieval tasks and goals.

Story Semantic Relationships from Multimodal Cognitions

“…The HLVU dataset (Table 1) has 10 open source movies sampled from paper [4]. The training set includes four long and two short movies, while testing set includes two long and two short movies.…”

“…The entities provided in the HLVU dataset [4] include person, objects, locations and concepts for which relevant images are provided for mapping. The locations and object entities are localized within scenes using SIFT based feature matching to handle varying scales and crops.…”

Kinetics and Scene Features for Intent Detection

“…The High-Level Video Understanding (HLVU) dataset [4] includes 10 movies that are suitable for researching the relationship between Figure 1: Architecture of the multi-modal fusion model entities. The HLVU dataset meets the important requirements for selecting movies such as the duration of the movies (different lengths of movies: 6 long movies and 4 short movies), the quality of the video, and the clarity of the storyline.…”

Towards Using Semantic-Web Technologies for Multi-Modal Knowledge Graph Construction

“…The challenge uses the recently introduced High Level Video Understanding (HLVU) dataset [5] which consists of 10 movies released under creative commons licenses with a total combined duration of 681 minutes. For each of the 10 videos, the dataset also provides cropped key-frames, showing either characters or locations which are relevant for the story told by the movies.…”

“…mechanism. 5 Detected faces are then compared against the provided example images in order to identify the visible person. Due to changes in size and orientation of the people on screen, as well as variations of overall image quality throughout the videos, we apply the face detection and identification method densely, i.e., on every frame of the videos.…”