search

tesseract-ocr / tesstrain

Train Tesseract LSTM with make
Apache License 2.0
620 stars 181 forks source link

From stweil's custom makefile - calculate MAX_ITERATIONS from EPOCHS #223

Closed Shreeshrii closed 3 years ago

Shreeshrii commented 3 years ago

See https://github.com/tesseract-ocr/tesstrain/issues/157#issuecomment-615873731 for the patch

stweil commented 3 years ago

I did not suggest that modification myself because it requires an existing list.train. That means it is not a general solution, and it is very likely that people will get unexpected results.

So we need a better solution to really support epochs. Maybe all programs which use the iteration parameter (lstmtraining, others?) can be enhanced to allow either a number (for iterations) or a number followed by epochs (for epochs).

Shreeshrii commented 3 years ago

Even if not merged, keeping this code in the PR makes it easier to find :-)

zdenop commented 3 years ago

Please use wiki for taking notes not issue tracker or PR.

stweil commented 3 years ago

Pull request https://github.com/tesseract-ocr/tesseract/pull/3236 allows a more stable implementation. With that change, we either need only an update of the tesstrain documentation because a negative value for MAX_ITERATIONS is then interpreted as epochs. Or we can add an EPOCH parameter which is forwarded to a negative MAX_ITERATIONS.

Shreeshrii commented 3 years ago

we can add an EPOCH parameter which is forwarded to a negative MAX_ITERATIONS.

I would prefer this option. Thanks!

Shreeshrii commented 3 years ago

Closing, in favor of https://github.com/tesseract-ocr/tesstrain/pull/228