4 ms·
> If you really can't afford mistakes you have to consider the OCR inaccurate. Isn’t this close to the error rate of human transcription for messy input, thoug
by kergonath 9mo ago
> If you really can't afford mistakes you have to consider the OCR inaccurate.
Isn’t this close to the error rate of human transcription for messy input, though? I seem to remember a figure in that ballpark. I think if your use case is this sensitive, then any transcription is suspicious.
- aliljet 9mo agoThis is precisely the real question. If you're exceeding human transcription, you may be generally pretty good. The question is what happens when you tell a human to become surgical about some part of the document, how then does the comparison change..