Research

2D and 3D Scene Understanding

My current research focus is on generic object recognition and deep learning for camera-based 3D scene understanding in complex urban environments. A key aspect includes automatic label generation to enable scalable systems.

Domain Adaptation and 3D Data Generation

The main research during my PhD involved domain adaptation for LiDAR-based perception systems. Rather than re-labeling datasets for a variety of domains, the aim was to re-use data from other domains even if no labels were available.

Semantic segmentation of LiDAR point clouds.

Realism estimation of local regions in point clouds.

Upsampling of LiDAR data.

Point cloud generation with continuous object descriptions.