convolutional neural networks (CNNs) can model complicated non-linear
relations between images. However, they are notoriously sensitive to small
changes in the input. Most CNNs trained to describe image-to-image mappings
generate temporally unstable results when applied to video sequen