example10_3_segmentation_custom_metric

from keras.models import Sequential
from keras.layers import Conv2D, MaxPool2D, UpSampling2D
from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import ModelCheckpoint
from keras import backend as K
import matplotlib.pyplot as plt

#Colab Version: https://colab.research.google.com/drive/1N4SrDU8hDiL6XTCCwqokWZagKvEy3n1m?usp=sharing

BATCH_SIZE = 5
MAX_EPOCH = 20
IMAGE_SIZE = (256,256)
TRAIN_IM = 160
VALIDATE_IM = 15

model = Sequential()
model.add(Conv2D(16, 3, activation='relu', padding='same', kernel_initializer='he_normal',
                 input_shape=(IMAGE_SIZE[0],IMAGE_SIZE[1],3)))
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Conv2D(32, 3, activation='relu', padding='same', kernel_initializer='he_normal'))
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Conv2D(64, 3, activation='relu', padding='same', kernel_initializer='he_normal'))
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Conv2D(128, 3, activation='relu', padding='same', kernel_initializer='he_normal'))
model.add(Conv2D(128, 3, activation='relu', padding='same', kernel_initializer='he_normal'))
model.add(UpSampling2D(size=(2, 2)))
model.add(Conv2D(64, 3, activation='relu', padding='same', kernel_initializer='he_normal'))
model.add(UpSampling2D(size=(2, 2)))
model.add(Conv2D(32, 3, activation='relu', padding='same', kernel_initializer='he_normal'))
model.add(UpSampling2D(size=(2, 2)))
model.add(Conv2D(16, 3, activation='relu', padding='same', kernel_initializer='he_normal'))
model.add(Conv2D(1, 3, activation='sigmoid', padding='same', kernel_initializer='he_normal'))

print(model.summary())


def iou(y_true, y_pred):
    y_true = K.cast(K.greater(y_true, 0.5), dtype='float32')
    y_pred = K.cast(K.greater(y_pred, 0.5), dtype='float32')
    inter = K.sum(K.sum(K.squeeze(y_true * y_pred, axis=3), axis=2), axis=1)
    union = K.sum(K.sum(K.squeeze(K.clip(y_true + y_pred, 0, 1), axis=3), axis=2), axis=1)
    return K.mean((inter + K.epsilon()) / (union + K.epsilon()))

model.compile(optimizer= 'adam', loss='binary_crossentropy', metrics=['accuracy',iou])

def myGenerator(type):
    datagen = ImageDataGenerator(rescale=1./255)

    input_generator = datagen.flow_from_directory(
        '/textlocalize/'+type,
        classes = ['Input'],
        class_mode=None,
        color_mode='rgb',
        target_size=IMAGE_SIZE,
        batch_size=BATCH_SIZE,
        shuffle=True,
        seed = 1)

    expected_output_generator = datagen.flow_from_directory(
        '/textlocalize/'+type,
        classes = ['Output'],
        class_mode=None,
        color_mode='grayscale',
        target_size=IMAGE_SIZE,
        batch_size=BATCH_SIZE,
        shuffle=True,
        seed = 1)

    while True:
        in_batch = input_generator.next()
        out_batch = expected_output_generator.next()
        yield in_batch, out_batch

checkpoint = ModelCheckpoint('my_model.h5', verbose=1, monitor='val_iou',save_best_only=True, mode='max')


h = model.fit_generator(myGenerator('train'),
                        steps_per_epoch=TRAIN_IM/BATCH_SIZE,
                        epochs=MAX_EPOCH,
                        validation_data=myGenerator('validation'),
                        validation_steps=VALIDATE_IM/BATCH_SIZE,
                        callbacks=[checkpoint])

plt.plot(h.history['iou'])
plt.plot(h.history['val_iou'])
plt.show()