Proposal-Free Instance Segmentation Algorithms and Applications for Autonomous Vehicles

scene understanding and environment perception are two important aspects for autonomous navigation. As both the number and quality of objects to be detected are increasing, the focus has shifted from traditional handcrafted feature extractors to data-driven neural networks, like Convolutional Neural Networks (CNNs). As a downside, CNNs require significant computational performance for both training and inference. During this PhD, the focus will be on the implementation of efficient and real time models, which can be used in realistic environments.