angelodel80 / printed-vs-handwritten Goto Github PK

System for classifying segmented text lines as printed or handwritten.
The provided neural network has been trained on:
* a private dataset for handwritten text lines
  + 9577 training lines
  + 1562 validation lines
* a dataset of printed text which will be released before August 2019
  + its address will be added here - if not, please send a reminder
  + 40459 training lines
  + 1562 test lines
The accuracy of the classifier on the validation data is 97.47%.

If you intend to use it, it is highly advised that you train it on your
own data. To do so, create the following folders:
  lines/training/handwritten
  lines/training/printed
  lines/validation/handwritten
  lines/validation/printed
and place your line images in these folders. Train the model by
running train-model.py, and evaluate it with evaluate-model.py.

Usage example:
python3 classify.py 616.jpg