Direkt zum InhaltDirekt zur SucheDirekt zur Navigation
▼ Zielgruppen ▼

Humboldt-Universität zu Berlin - Mathematisch-Naturwissenschaftliche Fakultät - Institut für Informatik

Institutskolloquium: Frau Dr. Dagmar Kainmüller

Wann 12.02.2020 ab 09:15 (Europe/Berlin / UTC100) iCal
Wo Rudower Chaussee 25, Humboldt-Kabinett
Kontaktname

Am Mittwoch, den 12.2.2020, wird um 9.00 Uhr c.t. Frau Dr. Dagmar Kainmüller, Nachwuchsgruppenleitein am MDC, im HUK einen Vortrag zum Thema

    Learning Instance Segmentation in Biomedical Image Data

halten.

Alle Institusmiglieder sind herzlich eingeladen.

 

Abstract 

Instance segmentation is the task of separately delineating each object in an image. It has a wide range of applications in natural images as well as in the biomedical domain, where there is, e.g., a pervasive need to delineate each cell of a biological sample in microscopy images. In this talk I will present our new method for instance segmentation, termed PatchPerPix, that outperforms the state of the art on benchmark microscopy data. Furthermore, I will discuss the related task of semantic parts segmentation, where instances are not just delineated, but also assigned individual semantic labels. I will present our efforts towards unsupervised semantic parts segmentation on the example of the worm C. elegans, a popular model organism that is stereotypically composed of a fixed number of cells of which each one has a unique biological name.