Pierre Jacob presents Image Representation using Metric Learning

On 2020-07-20 14:00:00 at G205, Karlovo náměstí 13, Praha 2
Learning rich and compact representations is an open topic in many computer
vision tasks, such as object recognition or image retrieval. Deep neural
networks have made a major breakthrough during the last few years, but the
representations are not as rich as needed nor compact as expected. In this talk,
we will discuss our contributions following two directions: the first one
addresses richer global pooling strategies that leverage attention mechanisms or
high-order statistics as well as dictionary learning, and the methods needed to
make them compact. The second one is focused on improving the training procedure
using a regularization method, and a strategy of hard example generation.

Bio: Pierre Jacob is a Ph.D. candidate at ETIS (Cergy-Paris University, ENSEA)
since March 2017 in the multimedia indexing group. He obtained his MSc in
Computer Science from Cergy-Pontoise University (Paris Area, France) and his MSc
in Electronics, Electrotechnics, and Computer Science from ENSEA (Paris Area,
France), both in September 2016. His research interests include image
representation learning with a focus on supervised metric learning, high-order
statistics, and attention mechanisms.
Za obsah zodpovídá: Petr Pošík