Please read TrainingTesseract 4.00
There have been many changes made to LSTM training process.
You can see the following links where there are modified training scripts created by Tesseract users:
- By J Klein at pastebin
- wiki.wareya.moe - info
- wiki.wareya.moe - tesstrain.sh at pastebin
- wiki.wareya.moe - tesstrain_utils.sh at pastebin
To train from line images and its matching ground truth, please see the project ocr-d/train which creates box files and lstmf files using the line images.