When set in the config files, it does not end up in the Dict where it is needed.
Probably because tesseract has the nasty habit of (re)initializing by ditching all subordinate class instances, one of which is the LSTMRecognizer, which carries the Dict.
Anyway, I guess this is an open problem until I get the Parameter Snapshots functionality done as that is meant to answer other questions and solve problems too.
Once you have a snapshot, then the idea is that a Dict instance being created should go to Tesseract and request the latest snapshot to fetch its parameter values from. (Needless to say Snapshots are meant to carry not-yet-known parameters as well as already-known-ones. 😉
GerHobbelt
commented
3 months ago
When set in the config files, it does not end up in the Dict where it is needed.
Probably because tesseract has the nasty habit of (re)initializing by ditching all subordinate class instances, one of which is the LSTMRecognizer, which carries the Dict.
Anyway, I guess this is an open problem until I get the Parameter Snapshots functionality done as that is meant to answer other questions and solve problems too. Once you have a snapshot, then the idea is that a Dict instance being created should go to Tesseract and request the latest snapshot to fetch its parameter values from. (Needless to say Snapshots are meant to carry not-yet-known parameters as well as already-known-ones. 😉