Detail of the student project

Topic:[Deep Learning] Stereo with Binary Descriptors
Department:Katedra kybernetiky
Supervisor:Mgr. Oleksandr Shekhovtsov, Ph.D.
Announce as:Diplomová práce, Bakalářská práce, Semestrální projekt
Description:Stereo reconstruction is based on finding correspondence between left and right images. This is a classical problem in computer vision with established benchmarks and very accurate methods available.
In this project you will design and train neural-networks for stereo reconstruction, that compute and compare local binary descriptors in the two images. This would allow to solve the problem faster, with less energy resources, and can scale better to more general correspondence search methods.
You will learn techniques for training with binary variables, design architectures for matching and formulations for training, derive improved statistical models.
Responsible person: Petr Pošík