On the synergies between machine learning and stereo

Matteo Poggi (University of Bologna, Italy)


Stereo matching is one of the longest-standing problems in computer vision and boasts over 30 years of studies and research. Although often in competition with active depth sensors and other image-based techniques, the advent of machine learning and the rapid spread of deep learning since 2014 rejuvenated this field with new exciting trends and applications unthinkable until a few years ago. Interestingly, the relationship between these two worlds is two-way. While machine/deep learning notably pushed forward the state-of-the-art of stereo matching, stereo enabled new ground-breaking methodologies such as self-supervised monocular depth. In this talk, we will give an overview of recent works leveraging on the synergies between the two, opening new research directions in the context of self-