Exploiting synthetic data to improve human behaviour understanding
Most recent Deep Learning techniques require large volumes of training data in order to achieve human-like performance. Especially in Computer Vision, datasets are expensive to create because they usually require a considerable manual effort that can not be automated. Indeed, manual annotation is error-prone, inconsistent for subjective tasks (e.g. age classification), and not applicable to particular data (e.g. high frame-rate videos).
For some tasks, like pose estimation and tracking, an alternative to manual annotation implies the use of wearable sensors. However, this approach is not feasible under some circumstances (e.g. in crowded scenarios) since the need to wear sensors limits its application to controlled environments. To overcome all the aforementioned limitations, we collected a set of synthetic datasets exploiting a photorealistic videogame. By relying on a virtual simulator, the annotations are error-free and always consistent as there is no manual annotation involved.
Matteo Fabbri: Postdoctoral Researcher at UNIMORE, CEO and cofounder at GoatAI
Matteo Fabbri pursued his PhD at the University of Modena and Reggio Emilia under the supervision of Prof. Rita Cucchiara. He is currently a postdoctoral researcher at the University of Modena and Reggio Emilia and CEO of GoatAI. He spent 10 months in Silicon Valley working as a deep learning engineer for Panasonic. His research interests include 3D pose estimation, multiple people tracking, and synthetic datasets.