GET-Forschungsseminar Abstracts
Semi-supervised Learning of View-based Object Representations
Markus Hennig, GET Lab
Vortrag: Mi. 23.04.2014, 16:30, Raum P 1.4.17
Abstract:
Robust and efficient object representations are required to practically analyse the status (presence, location, size, etc.) of objects within images or image sequences and therefore fundamental to detection and tracking algorithms. In contrast to object-centered representations, view-based object representations employ a set of views as observed by an external observer or camera without exploiting complex parts-based representations of shape. This talk focuses on approaches for automatic learning of such view-based representations using only a few labeled examples (semi- or weakly supervised learning and therefore without an extensive previous training phase). This allows for interacting with a range of further algorithms such as artificial visual attention algorithms. In particular, biological backgrounds, existing algorithms and approaches are evaluated and design criteria for future algorithms are identified.