Issue 1353: Patch for /training/tessopt.cpp #13
Conversation
https://code.google.com/p/tesseract-ocr/issues/detail?id=1353 When training batch boxes and tifs this causes training to fail after 2nd iteration. The issue is that the tessopt iterator index is a global. The fix is to 1.) reset before each tessopt usage 2.) wrap the iterator state in a class/generator. Low priority, but useful if we want to generate a bunch of traineddatas in the same process or even multiple threads.
|
is this PR relevant for 4.0? Any way it just implement tessopt_reset function but never used it... |
|
It is a 3 year old PR. I will request @stweil to review. Thanks. |
|
It looks like the start of a bug fix. Please keep it open as a reminder. I added the issue / pull request to the planning page in the wiki. |
|
I think the OP is @oulrich1 |
|
Wow, it's been a while. Seems a lot has changed since then. Doesn't look like 'tessopt' is used in 'lstmtraining', but still definitely used in 'cntraining' and maybe 'mftraining'. |
|
Closing as this it not (at least full) patch, there is no test case for replicating problem, so we can try to fix it... |
https://code.google.com/p/tesseract-ocr/issues/detail?id=1353
When training batch boxes and tifs this causes training to fail after 2nd iteration. The issue is that the tessopt iterator index is a global. The fix is to 1.) reset before each tessopt usage 2.) wrap the iterator state in a class/generator. Low priority, but useful if we want to generate a bunch of traineddatas in the same process or even multiple threads.
(http://web.archive.org/web/20150413012203/https://code.google.com/p/tesseract-ocr/issues/detail?id=1353)