OCR Training

This repository holds the script to train the neural network used in WhiteboadLiveCoding. It also houses the script to add images to the dataset.

Usage

training.py

usage: training.py --emnist <path-to-emnist.mat> --wlc <path-to-wlc.mat>

Required Arguments:

--emnist EMNIST       path for the EMNIST dataset
--wlc WLC             path for the WLC dataset

Optional Arguments

-h, --help            show the help message and exit
-o OUTPUT, --output OUTPUT
                      output directory for the model(without /)
                      default: bin
--height HEIGHT       height of the input image
                      default: 28
--width WIDTH         width of the input image
                      default: 28
-e EPOCHS, --epochs EPOCHS
                      number of epochs to train the model on
                      default: 10
-g GPUS, --gpus GPUS  number of gpus to be used
                      default: 1
-b BATCH, --batch BATCH
                      batch size for training
                      default: 64
-d DEVICE, --device DEVICE
                    device to be used for training
                    default: /cpu:0
-m MODEL, --model MODEL
                      keras model to be trained
                      default: convolutional
-p , --parallel       use multi gpu model
-f, --fix-emnist      fix the images from emnist

dataset.py

A Dataset class has been defined to load the wlc-byclass dataset and add more images to it.

    # Reference Code
    from OCRTraining.dataset import Dataset

    dataset = Dataset(batch_size=32) # The new images are cached and put in the dataset in a batch

    # This adds an image to the training data with label '{'
    dataset.add_image([for i in range(784)], '{')

    # This adds an image to the testing data with label '{'
    dataset.add_image([for i in range(784)], '{', test_data=True)

    # This saves the dataset to a .mat file with the timestamp as a part of the filename
    dataset.save()

    # A do_compression=False argument can be supplied to save an uncompressed dataset
    dataset.save(do_compression=False)

Based on a project by @coopss