jsmolka / tensor-ocr

Optical character recognition using Tensorflow.

tensor-ocr

Optical character recognition using Tensorflow.

Setup

1. Install Python 3.6 or lower (Tensorflow does not support newer versions).
2. Clone or download the repository.
3. Install the requirements using pip install -r requirements.txt.
4. Download a pretrained model from the release page if you do not want to train the model yourself.
5. Extract the archive and move the model.json and weights.h5 files into the data directory.

Training

If you want to train a model yourself you need to download the IAM dataset. After that you need to convert the dataset and train the model using the commands from the section below.

Commands

Command	Action
python app.pyw	Starts the GUI application.
python main.py --analyse	Analyses the IAM dataset.
python main.py --convert	Converts the IAM dataset into the desired format.
python main.py --rotate	Rotates the converted IAM dataset to create more training data.
python main.py --train	Trains the model using the converted IAM dataset.
python main.py --test	Tests the trained model.