If you are interested in how it compares to bicubic or the original. Check these papers using the sub pixel convolutional layer: https://arxiv.org/abs/1609.05158 https://arxiv.org/abs/1609.04802.