Recorded at the ML in PL 2019 Conference, the University of Warsaw, 22-24 November 2019.
Gül Varol (University of Oxford)
Slides available at docs.mlinpl.org...
Abstract:
This talk will present some recent works on human understanding from images and videos, with an emphasis on the use of synthetic data for training deep neural networks. Synthetic data is appealing due to providing free annotations and it has been shown beneficial for a set of tasks in computer vision. This talk will concentrate on human body shape estimation and hand-object reconstruction from images, as well as human action recognition from videos. I will briefly present approaches for addressing each of these tasks while illustrating the advantages of synthetic data obtained by rendering 3D body models.
29 сен 2024