//=time() ?>
currently on a little research binge exploring transfer learning vs from scratch in sg2-ada
so far i think it may still be true that transfer learning is better, but that theres a false hope that you can really train it in that much shorter of a time. still takes weeks.
me and @odomojuli are building some cool opencv tools to help create datasets. I never thought I’d be able to separate all these flowers without doing it by hand. Now its more sorting that lassoing so its prob like 2x efficient.
Add some new flowers, including my avatar at the low price of $35 plus shipping
https://t.co/HjAlLnyzNQ
As others have noted, I am bad at talking about my print shop. So here’s some new things in the store.
First some weird, abstract blob things:
https://t.co/Hc2IjA8NPR
Of course RUSH breaks the detector.
R U S H .
(beautiful little piece of typography tho)
working pretty well! (some issues with the photography in many sets but this will still probably 2-4x my dataset efficiency)
@pbaylies This isn’t the best example, but imagine a script that can find the bounding box of the drawing, can straighten it to 90degrees and crop down to it.
I have a couple illustrations in HP’s Innovation magazine. You can check out the digital issue here: https://t.co/aNH6jhVP6r