2 ms·

It absolutely hasn't been solved, it's just got pretty decent in recent years.

• malfist 3mo ago
  Pretty decent might be quiet the stretch. I'd term it almost acceptable, but only if you're using commercial solutions like amazon's textract, doing it with open source tools is at best, extremely painful and vaguely accurate.
  • chpatrick 3mo ago
    PaddleOCR (also from Baidu) is pretty damn good actually.
    • __rito__ 3mo ago
      I have shipped with PaddleOCR to prod. Works pretty well. (Usage limited to printed documents in Anglosphere). Runs fully offline, in CPU.