Michael Hirsch

Mehdi S. M. Sajjadi, Bernhard Schölkopf, Michael Hirsch

Single image super-resolution is the task of inferring a high-resolution image from a single low-resolution input. Traditionally, the performance of algorithms for this task is measured using pixel-wise reconstruction measures such as peak signal-to-noise ratio (PSNR) which have been shown to correlate poorly with the human perception of image quality. As a result, algorithms minimizing these metrics tend to produce over-smoothed images that lack high-frequency textures and do not look natural despite yielding high PSNR values. We propose a novel application of automated texture synthesis in combination with a perceptual loss focusing on creating realistic textures rather than optimizing for a pixel-accurate reproduction of ground truth images during training. By using feed-forward fully convolutional neural networks in an adversarial training setting, we achieve a significant boost in image quality at high magnification ratios. Extensive experiments on a number of datasets show the effectiveness of our approach, yielding state-of-the-art results in both quantitative and qualitative benchmarks.

Carsten Kolassa, David Dieckow, Michael Hirsch et al.

In software and hardware development MATLAB and Simulink are used to model cyber physical systems for many years, , especially in automation technology and the automotive industry. Compliance with the required product quality and project efficiency is facilitated by analyzing and transforming Simulink models. The existing API, provided by MATLAB is only suitable for programmatic changing of Simulink models. We show using our own tool which is used in industry, how such as a Simulink model can be edited more easily. For this purpose the model, is converted to an object-oriented class structure that provides convenient access and editing APIs and allows applying well-known algorithms and analyses from graph theory directly. It is also designed as a bi-directional tool, so it transforms a Simulink model into a graph representation and vice versa. ----- In der Software- und Hardwareentwicklung wird seit Jahren verstärkt MATLAB und Simulink für die Modellierung von cyberphysikalischen Systemen, insbesondere in der Automatisierungstechnik und der Automobilindustrie eingesetzt. Die Einhaltung der notwendigen Produktqualität und Projekteffizienz wird durch Analysen und Transformationen auf Simulink-Modellen erleichtert. Die bestehende, von MATLAB bereitgestellte, API ist für die programmatische Veränderung von Simulink-Modellen nur bedingt geeignet. Wir zeigen deshalb anhand eines eigenen, im industriellen Einsatz befindlichen Werkzeugs, wie ein Simulink-Modell leichter bearbeitet werden kann. Dazu wird es in eine objektorientierte Klassenstruktur überführt, die einen komfortablen Zugang und Bearbeitungs-APIs bietet und es erlaubt bekannte Algorithmen und Analysen aus der Graphentheorie direkt anzuwenden. Das Werkzeug ist bidirektional entworfen, es transformiert also ein Simulink-Modell in eine Graphenrepresentation und umgekehrt.

Christian J. Schuler, Michael Hirsch, Stefan Harmeling et al.

We describe a learning-based approach to blind image deconvolution. It uses a deep layered architecture, parts of which are borrowed from recent work on neural network learning, and parts of which incorporate computations that are specific to image deconvolution. The system is trained end-to-end on a set of artificially generated training examples, enabling competitive performance in blind deconvolution, both with respect to quality and runtime.