Project for the course "Industrial image processing" at Ostfalia University Wolfenbüttel in the winter term 2020/2021.
Activate the conda environment:
conda env create -f env.yml
conda env activate ibv-ocr
This requires the presence of the IAM dataset. You can download it at: https://fki.tic.heia-fr.ch/databases/iam-handwriting-database
Download words.tgz
and extract it into the data
folder.
python train.py --model_path=model --learning_rate=0.001 --augment=true --epochs=100 --batch_size=150
On the first run, all images are loaded into a lmdb for faster access times.
python htr.py model data\characters.txt examples\are.png
# -> are
python htr.py model data\characters.txt examples\bild.png
# -> Bild
python htr.py model data\characters.txt examples\wind.png
# -> wind (LER 0.25 w instead of W)
python htr.py model data\characters.txt examples\wort.png
# -> work (LER 0.25 k instead of t)
Start the gui with
python main.py
python htr.py model data\characters.txt examples\wind.png