there's another project that does this: https://erikbern.com/2016/01/21/analyzing-50k-fonts-using-deep-neural-networks.html https://erikbern.com/2016/01/21/anal

there's another project that does this: https://erikbern.com/2016/01/21/analyzing-50k-fonts-using-deep-neural-networks.html https://erikbern.com/2016/01/21/analyzing-50k-fonts-using-de... the results are a bit blurry but as the author notes the next step is to use a better loss function. A GAN loss or GAN+L1 ala pix2pix should dramatically improve results.