The One Where They Reconstructed 3D Humans and Environments in TV Shows

TV shows depict a wide variety of human behaviors and have been studied extensively for their potential to be rich source data many applications. However, the majority existing work focuses on 2D recognition tasks. In this paper, we make observation that there is certain persistence in shows, i.e., repetition environments humans, which makes possible 3D reconstruction content. Building insight, propose an automatic approach operates entire season show aggregates information 3D; build model environment, compute camera information, static scene structure body scale information. Then, demonstrate how acts as context can guide improve recovery pose position these environments. Moreover, reasoning about humans environment enables broad range downstream applications: re-identification, gaze estimation, cinematography image editing. We apply our from seven iconic perform extensive evaluation proposed system.