In recent years, the task of video prediction-forecasting future video given
past video frames-has attracted attention in the research community. In this
paper we propose a novel approach to this problem with Vector Quantized
Variational AutoEncoders (VQ-VAE). With VQ-VAE we compress h