Reading old county records: OCR plus a vision model
- Understandnow
- Explore
- Converge
- Commit
Next stepReid runs the OCR on the Ector County records and compares with Title Lab
StatusReid's chosen setup misreads about 5% of words, against 27% for ours; next steps after reading not planned yet
Waiting onWaiting on Reid
Last movedOct 26
ProjectNone yet
To-dos 2
Once Title Lab access comes through, compare how it handles the data with ours
Reid · from Oct 26 · Joe, Reid
Run the OCR on the Ector County records
Reid · from Oct 26 · Joe, Reid
Decisions 2
D18 Reid goes ahead with the OCR run on the Ector County records now, because Joe looked over the plan and is fine with it, and Keegan is fine paying for it to see what it can do
Agreed by Joe, Reid · accepted · from Oct 26 · Joe, Reid
D15 The Martin and Ector County records are a test of how far we can get; they do not change the hardware choice for now, because Reid does not want to build for a problem we may not have, and the first run may be the most expensive one
Agreed by Joe, Reid · confirmed by joe on the page · from Oct 26 · Joe, Reid
No questions, requirements, drafts or files yet.
History · 2
- Added to the board
- Reid tested about 20 ways to read scanned records. Words read wrong: his old setup 23%, ours (the one in our deal-data process) 27%, the setup he chose (OCR plus a vision model for handwriting and old pages) 5.1%. He wants it running in the cloud so no laptop has to stay on. What comes after reading, like tagging names, is not planned yet. His write-up is the document he shared.