There are even a couple of "Y"s in there, if we're being liberal 
-
-
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Shout out to
@_joelsimon and@noahtren by the way. Both have worked on generation using cooperation/communication over noisy channels and inspired this system with GlyphNet and Dimensions of DialoguePrikaži ovu nitHvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I'm really interested in this sort of cooperative generation! What is the keyword I should probably search with?
-
Dimensions of Dialogue and GlyphNet are two examples of the idea :)
Kraj razgovora
Novi razgovor -
-
-
How about adding a constraint that you have to be able to draw the generated letters in one stroke? I think that could create more natural-looking letters.
-
I agree! Though I'm unsure of how it'd be done with this model.
@hardmaru has a great piece on generating very realistic handwriting using stroke data: http://blog.otoro.net/2015/12/12/handwriting-generation-demo-in-tensorflow/ … It would be interesting to see what a similar model would look like if it could use only 1 stroke
Kraj razgovora
Novi razgovor -
-
-
this is really neat, the rectangular size constraint also makes the results more more letter-like. would love to see the code

-
Thanks! I'll post the colab notebook. I've been tweaking it since for higher resolution, but I'll try to match the original parameters in the published copy to reproduce similar results.
Kraj razgovora
Novi razgovor -
-
-
Cool - what changes if any were made over the original?
-
I shared the colab notebook with my code, and I haven't looked closely at yours, but I believe there are hyperparameter differences and the addition of loss from both NNs passing images back & forth
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.