Estimating scene geometry and semantics is a prerequisite for visual systems to interact with our physical 3D world. Because they are largely intertwined, these two cues are better estimated jointly.
In this line of research, I will present some of our recent works that estimate the complete 3D information either leveraging 3D data with lightweight 2D backbones, or using monocular 2D images. Beyond the visible scene parts, we will see that additional insights can boost estimation of the occluded areas.

The presentation will be given in English and the slides will be in English.