I am interested in estimating 3D scene representations from multi-view video sequences. In particular, I focus on combining semantic segmentation, object detection and classification with 3D reconstruction using efficient inference methods.

I am working on including physical properties into the visual inference process. The main research question is how to make use of sophisticated forward processes in computer graphics systems to facilitate scene understanding and description.

I am interested in efficient inference methods for computer vision. What makes models stand out to allow fast inference and how push the computational burden towards training time? In particular, I am working on human pose estimation from single images, a challenging structured prediction problem.

How can autonomous perception discover high-dimensional patterns in recorded data from our environment? I am approaching this question by working on structured computer vision tasks, such as Human Pose Estimation. I hope that insights from this area will improve our data analysis systems, so that they can assist us in better understanding our environment.

I'm working at the intersection of machine learning and computer vision, developing efficient non-parametric models of human pose, motion, appearance, and segmentation. I'm especially interested in exploiting the additional information provided by the dynamic context present in temporal data sources.

I work on decomposing photographs into intrinsic layers in order to do object detection and inpainting in images. Currently, I produce a new dataset for intrinsic image research, to be able to evaluate and compare intrinsic image algorithms on a broader range of data than publicly available at the moment.

My research focuses on the computational analysis of video sequences: In what ways can the temporal dimension of videos be used by a computer to better understand the structure of a scene? And what can we learn from dynamic stimuli processing in the human visual system to make our algorithms more robust?