Computational and Applied Mathematics seminar

Abstrakt finns enbart på engelska: Tensor decompositions are powerful tools for analyzing high-dimensional data by breaking down multi-way arrays into smaller, meaningful components. They help uncover patterns and handle missing data effectively. In this work, we consider two tensor decomposition methods, CANDECOMP/PARAFAC (CP) and Tucker, to address the problem of gap filling in motion capture (MoCap) data. The gap-filling problem in marker-based MoCap systems occurs when markers become occluded or detached during recording, resulting in incomplete data. Tensor decompositions offer an effective solution by leveraging the inherent multi-way structure of MoCap data. We develop and analyze different completion algorithms built upon CP and Tucker decompositions, and evaluate their performance across different missing data scenarios. The algorithms are tested using motion capture sequences from the publicly available CMU and HDM05 datasets.