Quantifying the Complexity of Standard Benchmarking Datasets for Long-Term Human Trajectory Prediction

Methods to quantify the complexity of trajectory datasets are still a missing\npiece in benchmarking human trajectory prediction models. In order to gain a\nbetter understanding of the complexity of trajectory prediction tasks and\nfollowing the intuition, that more complex datasets contain more information,\nan approach for quantifying the amount of information contained in a dataset\nfrom a prototype-based dataset representation is proposed. The dataset\nrepresentation is obtained by first employing a non-trivial spatial sequence\nalignment, which enables a subsequent learning vector quantization (LVQ) stage.\nA large-scale complexity analysis is conducted on several human trajectory\nprediction benchmarking datasets, followed by a brief discussion on indications\nfor human trajectory prediction and benchmarking.\n