I am David, a Computer Vision Researcher and PhD Candidate specializing in Data-Centric AI, with a specific focus on analyzing and mitigating label noise and systematic annotation uncertainty in 2D/3D scene understanding (perception tasks like object detection etc.). My research addresses the entire machine learning pipeline to deliver the foundational label-noise solutions required for robust machine perception in safety-critical industries such as autonomous driving, robotics, manufacturing, and medical imaging.
Key Contributions & Tooling:
- Methodology: Investigated how label noise and annotation contradictions introduce hidden performance limits, establishing the concept of “Label Convergence”. (WACV2025)
- Algorithmic Frameworks: Developed KaLOS, an open-source framework designed to standardize data quality metrics under the influence of spatial and categorical label noise. (CVPR2026)
- Open-Source Software: Programmed and published the Multi-Annotator Toolkit for the Voxel51/FiftyOne ecosystem to enable systematic visualization of annotation uncertainty and dataset flaws.
Reaching out to me:
- david.tschirschwitz@uni-weimar.de
- github.com/Madave94
- Google Scholar