(creator here) The copy is mostly referring to the text detection algorithm that was being used (It's built on a tweaked custom implementation of Microsoft Res

(creator here) The copy is mostly referring to the text detection algorithm that was being used (It's built on a tweaked custom implementation of Microsoft Research's Stroke Width Transform algorithm)— which was state of the art a few years ago (when I first wrote the words on the website). Nowadays neural approaches perform a bit better, so I should probably change that.