NOVA: Rendering Virtual Worlds with Humans for Computer Vision Tasks

Today, the cutting edge of computer vision research greatly depends on availability large datasets, which are critical for effectively training and testing new methods. Manually annotating visual data, however, is not only a labor-intensive process but also prone to errors. In this study, we present NOVA, versatile framework create realistic-looking 3D rendered worlds containing procedurally generated humans with rich pixel-level ground truth annotations. NOVA can simulate various environmental factors such as weather conditions or different times day, bring an exceptionally diverse set life, each having distinct body shape, gender age. To demonstrate NOVA's capabilities, generate two synthetic datasets person tracking. The first one includes 108 sequences, levels difficulty like tracking in crowded scenes at nighttime aims limits current state-of-the-art trackers. A second dataset 97 sequences normal used show how our be utilized train boost performance deep-learning based Our results indicate that data by represents good proxy real-world exploited tasks.