I am a PhD student in the Embodied Vision group and scholar in the Max Planck ETH Center for Learning Systems (CLS) under the joint supervision of Dr. Jörg Stückler and Prof. Dr. Marc Pollefeys.
My research interests lie in the broad field of computer vision and, more specifically, are currently focused on learning structured scene representations. I'm interested to learn about ways in which humans perceive their environment and how to use these ideas in combination with deep learning methods. Moreover, I want to explore options of combining this work with other tasks like segmentation or 3D reconstruction.
Before starting at the MPI-IS, I received my M.Sc. in Computer Science from the RWTH Aachen University, Germany. During this time, I worked as a student assistant at the Computer Vision Group on 3D object reconstruction from images and wrote my master thesis about '3D Instance Semantic Segmentation on Point Clouds' under the supervision of Prof. Dr. Bastian Leibe. Apart from this, I spent some time abroad during my Erasmus year at the Imperial College London and a research summer school at the Tokyo Institute of Technology.
Elich, C., Oswald, M. R., Pollefeys, M., Stueckler, J.
CoRR, abs/2010.04030, 2020 (article)
Representing scenes at the granularity of objects is a prerequisite for scene understanding and decision making. We propose a novel approach for learning multi-object 3D scene representations from images. A recurrent encoder regresses a latent representation of 3D shapes, poses and texture of each object from an input RGB image. The 3D shapes are represented continuously in function-space as signed distance functions (SDF) which we efficiently pre-train from example shapes in a supervised way. By differentiable rendering we then train our model to decompose scenes self-supervised from RGB-D images. Our approach learns to decompose images into the constituent objects of the scene and to infer their shape, pose and texture from a single view. We evaluate the accuracy of our model in inferring the 3D scene layout and demonstrate its generative capabilities.
Our goal is to understand the principles of Perception, Action and Learning in autonomous systems that successfully interact with complex environments and to use this understanding to design future systems