<ul><li>ProDiff is a new trajectory imputation framework that uses only two endpoints as minimal information.</li><li>It integrates prototype learning to embed human movement patterns and a denoising diffusion probabilistic model for robust spatiotemporal reconstruction.</li><li>Joint training with a tailored loss function ensures effective imputation, outperforming state-of-the-art methods by improving accuracy on different datasets.</li><li>Further analysis shows a high correlation between generated and real trajectories, indicating the effectiveness of the ProDiff approach.</li></ul>

ProDiff: Prototype-Guided Diffusion for Minimal Information Trajectory Imputation

Discover more