Archive

Semi-automatic Data Annotation for Machine Learning

This project, financed by the Balgrist Foundation, aims to develop an application that accelerates the generation of datasets required for the training of machine learning algorithms for object recognition.

Future methods of surgical navigation, such as holographic navigation with Augmented Reality (AR), require a semantic segmentation of the surgical procedure. In the semantic segmentation of camera images, objects are recognized and associated with each other based on their semantic characteristics. Deep learning methods are successfully used for the object recognition task. However, a prerequisite is that these methods learn the appearance of the objects using training data in which all objects of a video have been marked and annotated by a human. This time-consuming process is called "labelling".

The project, which is funded by the Balgrist Foundation, aims to develop a software that reduces the time needed for the generation of training data sets for machine learning.

The application will not only enable users to conveniently mark objects in videos (Fig. 1), it will learn during the usage and present suggestions to the user, which the user then only has to accept or adapt. The application will be based on the Microsoft graphics framework 'Windows Presentation Foundation' (WPF) and C#.

 

Project Team & Cooperations

Sara Bayer
Biomedical Engineer

Profil

 

Balgrist Foundation


 

Webseite