Detail of the student project

Topic:Learning multilayer classifiers with weak annotations
Department:Katedra kybernetiky
Supervisor:Prof. Dr. Ing. Jan Kybic
Announce as:DP,BP,PMI,SOP,PRO
Description:This topic is similar to deep learning but without neural networks. We chain several classifiers together - for example the first one will do a pixel-wise segmentations and the second will evaluate the shape of the detected objects. The challenge is to train all classifiers in the chain when training data is only available for the final output - a so-called weak labels. We shall start by simulating several simple cases, gradually moving towards a practical method.
Realization form:SW project, theoretical analysis
Responsible person: Petr Pošík