cnn_comp.py

import numpy as np
from keras.models import Sequential
from keras.layers import Input, Conv2D, Dense, Dropout, MaxPooling2D, Flatten
from keras.callbacks import Callback, ModelCheckpoint
from keras.utils import plot_model
from keras import backend as K
from utils import *

IMAGE_SIZE = get('image_dim')
DROPOUT_CNN = get('cnn.dropout_cnn')
DROPOUT_DENSE = get('cnn.dropout_dense')
OPTIMIZER = get('cnn.optimizer')

def generate_model():
	# define CNN
	model = Sequential()
	# (128,128,3)
	model.add(Conv2D(16, kernel_size=(7, 7),
	                 activation='relu',
	                 input_shape=(IMAGE_SIZE,IMAGE_SIZE,3)))
	# (122,122,16)
	model.add(Conv2D(32, (5, 5), activation='relu'))
	# (118,118,32)
	model.add(MaxPooling2D(pool_size=(2, 2)))
	# (59,59,32)
	model.add(Dropout(DROPOUT_CNN))
	# (59,59,32)
	model.add(Conv2D(64, (5, 5), activation='relu'))
	# (55,55,64)
	model.add(Conv2D(64, (3, 3), activation='relu'))
	# (53,53,64)
	model.add(MaxPooling2D(pool_size=(2, 2)))
	# (26,26,64)
	model.add(Dropout(DROPOUT_CNN))
	# (26,26,64)
	model.add(Conv2D(256, (3, 3), activation='relu'))
	# (24,24,256)
	model.add(Conv2D(256, (3, 3), activation='relu'))
	# (22,22,256)
	model.add(MaxPooling2D(pool_size=(2, 2)))
	# (11,11,256)
	model.add(Dropout(DROPOUT_CNN))
	# (11,11,256)
	model.add(Flatten())
	# 30976
	model.add(Dense(1250, activation='relu'))
	# 1250
	model.add(Dropout(DROPOUT_DENSE))
	# 1250
	model.add(Dense(1000, activation='relu'))
	# 1000
	model.add(Dropout(DROPOUT_DENSE))
	# 1000
	model.add(Dense(133, activation='softmax'))

	model.compile(loss='categorical_crossentropy',
	              optimizer=OPTIMIZER,
	              metrics=['accuracy'])

	plot_model(model, to_file='model-comp.png')
	return model