Master's thesis on enhancing the robustness of LiDAR-camera fusion for object detection in automotive applications, with a focus on mitigating the impact of individual sensor failures.

Camera-based 3D scene reconstruction implementation inspired by "Diffeomorphic Neural Surface Parameterization for 3D and Reflectance Acquisition" by Cheng et al. (Paper, Code). We investigate the viability of this reconstruction approach for non-trivial backgrounds.

Implementation of graphics and shader processing in Vulkan for a 3D game engine. (since basic scaffolding was provided by TUM CGV, the full source is not publicly available)

Implementation of a 3D physics engine in C++ using OpenGL.

Autoregressive synthesis of realistic indoor scene layouts. An Autoencoder in voxel space allows for direct generation of the furniture, resulting efficient and diverse furniture arrangements.

We implement a voxel carving algorithm in C++ to reconstruct a colored 3D object from multiple RGB images, using pose estimation with ChArUco markers and image segmentation for preprocessing.

In this bachelor's thesis, we evaluate a pre-trained Scene Representation Network for modeling a smoke plume. We focus on analyzing the similarity of advected streamlines within the vector field.