Computer Graphics and Multimedia Systems

We investigate real-time RGB-D mapping using multiple complementary scene representations with adaptive mapping of geometry and/or appearance. Different representations offer specific advantages, for example, in terms of stable geometry, tracking, base color, high-frequency surface details, or view-dependent appearance. The goal is to progressively combine multiple representations, each with its own local advantages, to improve the quality of photorealistic novel views and the compactness of the map without compromising the speed and robustness of classical methods.

Our research focuses on the Material Point Method (MPM) as a powerful framework for continuous physical simulation. Our goal is to improve the computational efficiency of this method for large-scale simulations, extend it to a broader range of physical models and phenomena relevant to visual production, and develop novel applications in the fields of graphics and visual computing.

Our research focuses on real-time 3D reconstruction and the understanding of dynamic environments from a first-person perspective. We investigate methods that combine RGB-D cameras with inertial and eye-tracking data to achieve robust camera tracking, reconstruct static and moving objects, and capture human activities and interactions. The goal is to progressively integrate this multimodal information into a consistent spatiotemporal representation to understand scenes and activities in real time.

We are investigating how visualization can help us understand complex and often uncertain data—not only to recognize what the data reveals, but also to determine the extent to which these findings can be trusted. Therefore, we focus on interactive applications involving uncertainty modeling, propagation, and visualization, dimensionality reduction, and parameter space analysis, in order to combine computational methods with human strengths.