Behzad Sezari hat sein Masterstudium (ITIS) an der Leibniz Universität Hannover erfolgreich abgeschlossen. Das Thema seiner Masterarbeit lautet: „Cost-Effective Deep Active-Learning In Autonomous Driving Vehicles“.
The visual perception and understanding in the autonomous driving application is a subject to stringent performance requirement. Recently, incredible progress on visual recognition tasks has been made by deep learning approaches. With sufficient annotated data, deep convolutional neural networks (CNNs) are trained to directly learn features from raw pixels, which have resulted in very good performance for different tasks such as: image-classification, object-detection and semantic-segmentation. However, these approaches are data-driven and highly dependent on large number of annotated training samples, which may require considerable amount of time and human effort for producing labels for images. The goal of the master thesis of Mr. Sezari is to improve the current machine learning workflow by adapting an active-learning approach. It is conducted by intelligent selection of some portion of unlabeled data, annotating the most informative samples and still achieves promising performance by keeping human in the loop for instant quality control and labeling.