Robert Manthey

Annotate. Train. Evaluate. A Unified Tool for the Analysis and Visualization of Workflows in Machine Learning Applied to Object Detection

The development of classifiers for object detection in images is a complex task that comprises the creation of representative and potentially large datasets from a target object by repetitive and time-consuming intellectual annotations, followed by a sequence of methods to train, evaluate and optimize the generated classifier. This is conventionally achieved by the usage and combination of many different tools. Here, we present a holistic approach to this scenario by providing a unified tool that covers the single development stages in one solution to facilitate the development process. We prove this concept by the example of creating a face detection classifier.

A support framework for automated video and multimedia workflows for production and archive

The management of the massive amount of data in video- and multimedia workflows is a hard and expensive work that requires much personnel and technical resources. Our flexible and scalable open source middleware framework offers solution approaches for the automated handling of the ingest and the workflow by an automated acquisition of all available information. By using an XML format to describe the processes, we provide an easy, fast and well-priced solution without the need for specific human skills.

An Exploratory Comparison of the Visual Quality of Virtual Reality Systems Based on Device-Independent Testsets

Nowadays, several different devices exist to offer virtual, augmented and mixed reality to show artificial objects. Measurements of the quality or the correctness of their resulting visual structures are not developed as sophisticated as in the classical areas of 2D image and video processing. Common testsets for image and video processing frequently contain sequences from the real world to reproduce their intrinsic characteristics and properties as well as artificial structures to provoke potential visual errors (see Fig. 1a). These common but traditional testsets are nowadays faced with rapid technical developments and changes like HD, UHD etc. improved surround sound or multiple data streams. That results in a limitation of the testsets usability and their ability to evoke visual errors. To overcome those limitations, we developed a system to create device-independent testsets to be used in the area of virtual reality devices and 3D environments. We conduct an empirical evaluation of most recent virtual reality devices like HTC Vive and Zeiss Cinemizer OLED, aiming to explore whether the technical hardware properties of the devices or the provided software interfaces may introduce errors in the visual representation. The devices are going to be evaluated by a group with technical skills and mostly advanced knowledge in computer graphics. All perceived visual and technical saliences are recorded in order to evaluate the correctness and the quality of the devices and the constraints.

Simplifying Accessibility Without Data Loss: An Exploratory Study on Object Preserving Keyframe Culling

Our approach to multimedia big data is based on data reduction and processing techniques for the extraction of the most relevant information in form of instances of five different object classes selected from the TRECVid Evaluation campaign on a shot-level basis on 4 h of video footage from the BBC EastEnders series. In order to reduce the amount of data to be processed, we apply an adaptive extraction scheme that varies in the number of representative keyframes. Still, many duplicates of the scenery can be found. Within a cascaded exploratory study of four tasks, we show the opportunity to reduce the representative data, i.e. the number of extracted keyframes, by up to 84 % while maintaining more than 82 % of the appearing instances of object classes.