I’m working on Level 3 assignment 1, and chose OCR to scrape several pieces of information because the video instruction didn’t match up with the UI, and the emphasis on OCR in the course material led me to believe this was what I was supposed to do.
I’ve submitted and tuned things, but I’m not achieving good results. First attempt was ~7/12 (58/100) correct items, and the 2nd was ~8/12 (67/100). From what I can tell, the errors come down to OCR elements not being brought in correctly.
I noticed that the OCR was not pulling in the Client ID properly, so I tweaked that, eventually turning the scale up to 10. I caught it scanning a “D” as an O or 0, but tried again anyways.
Tessaract (Google) does better than Microsoft, but this level of performance is not what I would expect in 2019 in a commercial product where this is the backup, and the emphasis of much marketing hype, documentation, and training.
Is that just the way it is?