Field-of-view prediction in 360-degree videos with attention-based neural encoder-decoder networks.