Caching in TrainLineRecognizer?

[Jens Weibler]

Hi,

I'm new to tesseract and wondered why the lstm dataset creation for the training process has to write the file again and again in TrainLineRecognizer. I've seen 200MB/s IO on the disk while creating the training data set.

As far I can see for the training case it would be sufficient to just load it once and write it at the end. The same applies to the box and tif file - but these are only read and not written...

Thanks,

Jens Weibler