Institute for Data Processing and Electronics (IPE)
Area of research:
Recent improvements in detector instrumentation provide unprecedented details to researchers. At the same time the data rates are continuously increasing. It is a challenge to quickly and efficiently extract knowledge from the waste volumes of data and present it to users in easy to interpret visual form. Advanced visualization techniques are essential for collaboration in the international scientific community and to realize useful raw data catalogs. This is equally true for the high energy physics at LHC, planned future lepton and neutrino detectors, as well as for experiments at high-intensity light-sources such as the EU-XFEL or PETRA-III. Techniques based on ML promise to revolutionize data processing. If successfully deployed they have potential to realize an unprecedented complexity of algorithms and might offer significant improvements in the visualization of complex datasets for applications like X-ray tomography.
You are going to develop a novel framework for 3D data visualization relaying on classical and ML methods for perception enhancement. With a 3 stage visualization workflow combining pre-processing as well as a server- and client-side rendering we want to cater a high-quality visualization to clients ranging from handhelds to large visualization stations and keep the load on the server-side infrastructure to a minimum. The ML techniques should be evaluated and when applicable used along with traditional methods for data reduction, quality enhancements, and visualization. Early automated detection of non-standard and potentially problematic data may be enabled using ML techniques as well.
As a pilot project, an archive of samples produced at synchrotron facilities for research in developmental biology is chosen. The visualization framework is expected to show different aspects of the stored data, e.g. visualization of raw, pre-processed, and segmented data; multi-modal data visualization; visualization of uncertainty in segmentation; visualization of time-resolved (4D) tomographic volumes. Optimal data organization should be proposed to enable fast visualization of a region of interest. Existing traditional and ML-based methods should reviewed and optimal solution selected in order to prepare data for visualization. This includes correction of reconstruction artifacts, removal of holders/containers, selection of optimal initial view, noise reduction, etc. Further, the intelligent data reduction techniques are required. It is necessary to extract the reduced datasets suitable for visualization on the client hardware, but as much as possible representative of the original dataset.
You have a university degree (diploma (Uni) / Master) in the field of computer Science, mathematics or physics. You also have a good background in deep learning, image processing, and rendering. Familiarity with Python and a stack of relevant Python libraries. Experience in web development is a plus.
limited to 3 years
Application up to
Contact person in line-management