TVQA: Localized, Compositional Video Question Answering
This addresses the problem of limited data for video QA research, providing a new benchmark for the community, though it is incremental as it extends existing QA tasks to video.
The authors tackled the lack of video-based question-answering datasets by introducing TVQA, a large-scale dataset with 152,545 QA pairs from 460 hours of TV show clips, designed to require localized and compositional reasoning.
Recent years have witnessed an increasing interest in image-based question-answering (QA) tasks. However, due to data limitations, there has been much less work on video-based QA. In this paper, we present TVQA, a large-scale video QA dataset based on 6 popular TV shows. TVQA consists of 152,545 QA pairs from 21,793 clips, spanning over 460 hours of video. Questions are designed to be compositional in nature, requiring systems to jointly localize relevant moments within a clip, comprehend subtitle-based dialogue, and recognize relevant visual concepts. We provide analyses of this new dataset as well as several baselines and a multi-stream end-to-end trainable neural network framework for the TVQA task. The dataset is publicly available at http://tvqa.cs.unc.edu.