To make sure that the model is not overfitted to training data, your eval
set should be different.
You can use a different text file, different fonts from the training set to
check that the model performs well on text and fonts it has not seen
earlier.
On Tue 10 Apr, 2018, 8:16 PM Fanatico, wrot
Platform: MAC OS X
Tesseract: 4.0.0-beta.1-69-g10f4
Wen I execute a command like:
SCROLLVIEW_PATH=~/projects/tesseract/java \
~/projects/tesseract/training/lstmtraining \
--debug_interval 100 \
--continue_from ~/projects/ocr/training/kortrain/kor_from_full/kor.lstm
\
--traineddata
2 matches
Mail list logo