3 ms·

For the curious and lazy: it works by rendering the signed message to an image and tweeting that [2]. Then it relies on the fixed grid [0] and font to do a "po

• Lukasa 11y ago
  Entirely accurate! The interesting thing about the poor man's OCR is that it works way better than out of the box OCR tools. They're all built around handling arbitrary, often quite noisy input, so their handling of 1s and ls in this data is often not great. Doing the cheap OCR works really well because we can overfit our data: we know the font, we know the grid, and we know the orientation, so our problem is way simpler than generic OCR.