Loadingā¦
9
Old OCR scans from 2003 still stun me
Was digging through an archive of scanned documents and saw an OCR output from 2003 that got a 95% accuracy rate on messy handwriting, I honestly didn't think that tech existed back then. Does anyone else look at these old milestones and wonder if we're just adding flashy features now instead of real progress?
1 comments
Log in to join the discussion
Log In1 Comment
holly_craig17d ago
What kind of messy handwriting are we talking about here, because that detail matters a lot? Cursive from one writer who's consistent is a totally different problem than a doctor's scrawl or a form filled out by fifty different people. I remember reading that those old systems got decent numbers by being trained on very specific datasets, like postal codes or bank checks, where the range of shapes was narrow. So a 95% on that archive might say more about the test set than the tech itself. Do you know what the source documents actually looked like?
3