Personal Projects

Effective Negative Samples for Open-Set Classification - Master's Thesis

Deep neural network classifiers have a hard time recognizing what they do not know, which often leads to misclassifications and obvious mistakes. This problem is tackled by so called Open Set Classification approaches, which try to enable deep neural network classifiers to reject unknown data and correctly classify known samples. With the aim to effectively train these classifiers to distinguish between the known and unknown, artificially generated training data is created and used. My master thesis compares existing as well as explores new methods to generate such effective negative training data.

3D Point Cloud VR Viewer - Master Project

The Master Project involved developing a VR viewer for point clouds of indoor spaces (ASCII PTX files) using OpenVR and C++. An HTC Vive headset with two base stations was used for immersive visualization and precise movement tracking. Toggle navigation was added, allowing users to explore rooms by hovering over the floor. The source code for this project cannot be shared publicly.

Data Pipeline for Earthquake Visualization

I built a containerized, near-real-time data pipeline that fetches earthquake data from the USGS API using Python, stores it in PostgreSQL, and visualizes it with a Bokeh dashboard — all managed via Docker Compose.