GET-Forschungsseminar Abstracts
Semi-automatic dataset creation for a convolutional neural network
Bashar Tomeh, GET Lab
Vortrag: Mi. 13.06.2018, 16:30, Raum P 1.4.17
Abstract:
Generating training data for CNNs is a challenging and time-consuming task. In the RoboCup scenario, this is typically performed manually. This presentation focuses on a new approach to generate training data by automatic labeling of videos using an object tracker. A user interface for selecting good training examples further facilitates the process. The dataset was tested with a single shot detector and compared to an existing manually generated dataset.