train.py

import sys
from mjsynth_gen import MJSynthData
from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.optimizers import Adadelta
from crnn_model import get_model
from param import batch_size, validation_batch_size, epochs


def main(args):
    model = get_model(training=True)
    model.summary()
    # get weights to train
    try:
        model.load_weights(args[3])
        print("....Previous weight data...")
    except:
        print("...New weight data...")
        pass

    early = EarlyStopping(monitor='loss', min_delta=0.001,
                          patience=4, mode='auto', verbose=1)
    checkpoint = ModelCheckpoint(
        filepath='crnn_{epoch:02d}_{val_loss:.3f}.hdf5',
        monitor='loss',
        verbose=1,
        mode='auto',
        period=1)

    ada = Adadelta(rho=0.9)
    # sgd = SGD(lr=0.02, decay=1e-6, momentum=0.9, nesterov=True, clipnorm=5)
    model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer=ada)

    training_gen = MJSynthData(args[0], args[1], batch_size)
    validation_gen = MJSynthData(args[0], args[2], validation_batch_size)

    model.fit_generator(generator=training_gen,
                        steps_per_epoch=int(training_gen.dataset_size
                                            // batch_size),
                        epochs=epochs,
                        callbacks=[early, checkpoint],
                        validation_data=validation_gen,
                        validation_steps=int(validation_gen.dataset_size
                                             // validation_batch_size),
                        max_queue_size=32)


if __name__ == "__main__":
    if(len(sys.argv) != 5):
        print('Invalid command line inputs')
        exit()
    else:
        main(sys.argv[1:])