Readspeaker menu

M.Sc. Dimitri Korsch

Research Associate
Dimitri Korsch


Address: Computer Vision Group
Department of Mathematics and Computer Science
Friedrich Schiller University of Jena
Ernst-Abbe-Platz 2
07743 Jena
Phone: +49 (0) 3641 9 46426
E-mail: dimitri (dot) korsch (at) uni-jena (dot) de
Room: 1223

Curriculum Vitæ

Since October 2016 Research Associate at the Computer Vision Group of Friedrich Schiller University Jena
October 2013 September 2016 Master Degree in IT-Systems Engineering at Hasso-Plattner-Institute (University of Potsdam)
Master Thesis: "Rotation Estimation and Perspective Rectification of Scene Text"
December 2014 September 2016 Research assistant at Multimedia Analysis group of Hasso-Plattner-Institute. Focus: End-to-End Scene Text Recognition on Mobile Devices
October 2010 September 2013 Bachelor Degree in IT-Systems Engineering at Hasso-Plattner-Institute (University of Potsdam)
Bachelor Thesis: "Solving the Object-Relational Impedance Mismatch with a Persistent Programming Language"
Juni 2012 December 2013 Research assistant at Internet Security group of Hasso-Plattner-Institute. Focus: CloudRAID - Secure Storage in the Cloud [source code]

Research Interests

Part-based approaches for Fine-grained Visual Categorization

Fine-grained visual categorization is a classification task for distinguishing categories with high intra-class and small inter-class variance. While global approaches aim at using the whole image for performing the classification, part-based solutions gather additional local information in terms of attentions or parts. Hence, we research different part estimation approaches and part-classification methods.
Besides the part estimation and part-based classification approaches, we aim to go a step further and decide which of the estimated parts contribute the most to the final classification. We call this approach the Active Part Selection because the decision should be made actively based on previously known information, like initial classification result, previously selected parts or the uncertainty of the classifier. From our point of view, Active Part Selection separates into three different steps: part estimation, part-based classification, and the actual part selection process. Hence, our work focuses on these aspects.