search

mittagessen / kraken

OCR engine for all the languages
http://kraken.re
Apache License 2.0
750 stars 131 forks source link

Report missing after training #470

Closed kabikaj closed 7 months ago

kabikaj commented 1 year ago

Hi, I am trying to make a test for training the recognizer with several exported page xml documents done in escriptorium. The following command produces a list of models apparently without errors but no report on the accuracy:

$ ketos -vv train d cuda:0 --batch-size 16 -o out/model -f page export_page_format/*.xml

If I used the best model against an unseen image, I don't get any results. I guess that that must be because I need to produce more training data, but I don't understand why I don't get a report from the training. I have the feeling that I am doing something wrong

thanks! Alicia

colibrisson commented 1 year ago

Could you post the output of ketos train?

kabikaj commented 1 year ago

sorry for the late reply. I tried to upgrade pytorch and ketos after posting my comment and I then I started training again. I waited until it finished and now it works!, so I guess the problem was related with the versions.