With the rise of Deep Learning, paradigms have changed in several well-established fields of computer science. Algorithms that had been developed for decades were instantly outperformed by deep neural networks and often levels of performance were reached that seemed unimaginable. Computer Vision was one of the first areas that were conquered by Deep Learning. AlexNet, a deep Convolutional Neural Network for image classification achieved outstanding results in the 2012 ImageNet competition and ignited the Deep Learning hype and became a baseline for the following deep architectures.
While image classification finds a plethora of applications in i.e. web services, many practical applications depend not only on knowing whether a given pattern is present or not in images, but on knowing the patterns’ exact location. This challenge is addressed by object detection, the detection of objects’ bounding boxes, and semantic segmentation, the classification of every pixel of a given image. One major trait of semantic segmentation is that pixel-accurate class differentiation allows for precise analysis of an image’s composition.
This talk aims to highlight characteristics and promising applications of semantic segmentation. For this purpose, the relation between semantic segmentation and other current image recognition methods is investigated, state-of-the-art methods for semantic segmentation are discussed and suitable hardware and frameworks for training and deployment of semantic segmentation networks are addressed.
Felix Friedmann holds a diploma in Electrical Engineering from Technical University of Munich (TUM). Machine Learning was a strong focus of his studies at TUM. Next to his education, he worked for eight years as a part time software engineer and contributed to software projects in various industries. After graduating, he joined Audi Electronics Venture (AEV) GmbH in 2014, where he current works on the application of Deep Learning to autonomous driving. Felix is passionate about exploring new neural network architectures, deployment of AI on embedded platforms, and the manifold opportunities Deep Learning offers for automotive applications.