Stunning advance in Machine Learning applied to typeface design

Comments

  • No HTTPS? Unsubscribed, disliked, blocked, never touch again. I'm not even mad, just disappointed.

    Now with 100% less dumb: with a sufficiently good (read, accurate) model, could we maybe figure out what a script we have no familiarity with would look like, given certain design parameters, such as say a multi-script family?
    I guess for that to happen the model would need to be trained to have a similar opinion to a native specialist consultant, so the training data would have to come from those?
  • Sérgio Martins It's more a scientific  April fool hoax. Maybe you mean some serious attempts like this:

    Hideaki Hayashi; Kohtaro Abe; Seiichi Uchida; 
    GlyphGAN: Style-Consistent Font Generation Based on Generative Adversarial Networks
    https://arxiv.org/pdf/1905.12502.pdf

    Tomo Miyazaki, Tatsunori Tsuchiya, Yoshihiro Sugaya, Shinichiro Omachi, Masakazu Iwamura, Seiichi Uchida, and Koichi Kise;

    Automatic Generation of Typographic Font from a Small Font Subset


    Nikita Srivatsan, Jonathan T. Barron, Dan Klein, Taylor Berg-Kirkpatrick;
    A Deep Factorization of Style and Structure in Fonts
  • I didn't even see the SIGBOVIK… welp I feel stupid.

    I just assumed all the tongue in cheek humour about uppercaserer and lowercaserer (and the chess algorithms) were just youtube attention grabbing jokes.  :(
  • Dave Crossland Funny and entertaining.

    Brings some chaos into state of the art theory. Thus got some nice ideas like: Why not generate all possible combinations of 8 x 8 pixels and filter away the senseless (i. e. not matching a cognitive model of a character).


  • Mark SimonsonMark Simonson Posts: 1,381
    Interesting and funny. Laughed out loud a few times. ("Maybe a 2.5 on the Bristol stool scale?")
  • Craig EliasonCraig Eliason Posts: 1,179
    edited April 2

    Brings some chaos into state of the art theory. Thus got some nice ideas like: Why not generate all possible combinations of 8 x 8 pixels and filter away the senseless (i. e. not matching a cognitive model of a character).


    You may be interested in the work of Jurriaan Schrofer (whom I encountered through @Maurice Meilleur's research).
Sign In or Register to comment.