Aug 9, 2017 · 1 min read
The essense of what is going on can be considered analogous to autoencoders, but the presence of CO, which is the JPEG codec is the most fundamental difference. This algorithm doesnt learn representations that can recreate the original completely, but learns representations that allows the codec to recreate the image better. So in a way , this setup learns to work alongside a codec , and not simply on its own
