A tracking-based approach for video and volume annotation with - PowerPoint PPT Presentation

A tracking-based approach for video and volume annotation with sparse point supervision L. Lejeune, J. Grossrieder, R. Sznitman Ophthalmic Technology Laboratory, University of Bern, Switzerland

Problem • Segmentation relies more and more on complex Machine Learning models • Large amounts of ground truth annotations are necessary • Annotating video/volumetric sequences is tedious

Ambitions • Reduce user inputs to a minimum • No prior knowledge on the object of interest • Perform on a wide range of datasets Approach • User provides a single 2D location on the object of interest on each frame • Leverage frame-to-frame consistencies to propagate belief through a network Advantages • We can potentially annotate at frame rate .

A tracking-based approach for video and volume annotation with sparse point supervision L. Lejeune, J. Grossrieder, R. Sznitman Ophthalmic Technology Laboratory, University of Bern, Switzerland Try it out at www.gazelabel.com