tSPM-Net: A probabilistic spatio-temporal approach for scanpath prediction

Daniel Martin, Diego Gutierrez, Belén Masiá

Abstract: Highlights•We propose a probabilistic model for scanpath prediction.•Our model uses Bayesian deep learning and a probabilistic sampling strategy to select points in the scanpath.•We present a novel spatio-temporal loss function that combines Dynamic Time Warping and Kullback–Leibler Divergence.•Our model resorts to convolutional recurrent units to learn the spatio-temporal features of visual attention.•Our approach outperforms SOTA works for scanpath prediction.