The talk will introduce a system to automatically classify Drosophila embryos
into developmental stages.
In general, it will discuss how to take advantage of two orthogonal sources of
information, one that is globally valid but rather noisy and another that is
much stronger but only locally valid.
On the biological example, it will demonstrate that a combination of two
sources using label propagation provides an improved performance over any of
the two sources used individually.
Despite the biological motivation, the ideas behind the system might be of
interest to a broader computer vision community.
This talk is based on an ICCV 2013 paper:
T Kazmar, EZ Kvon, A Stark, CH Lampert: Drosophila embryo stage annotation using
About the speaker from Jan Kybic: Tomas Kazmar was my diploma thesis student
back in 2008. He got his master degree from MFF UK (Charles University). He is
now a PhD student in Vienna, jointly supervised by Alex Stark and Christoph