def train_and_assess_model(model, train_images, train_labels, test_images,
test_labels):
epochs = 50
learning_rate = 0.01
decay = learning_rate / epochs
sgd = SGD(lr=learning_rate, momentum=0.9, decay=decay, nesterov=False)
model.compile(loss='categorical_crossentropy',
optimizer=sgd,
metrics=['accuracy'])
print(model.summary())
model.fit(train_images,
train_labels,
validation_data=(train_images, train_labels),
epochs=epochs,
batch_size=12)
print(model.input.op.name)
print(model.output.op.name)
frozen_graph = freeze_graph(K.get_session(),
output_names=[model.output.op.name])
tf.train.write_graph(frozen_graph,
'.',
'emotions_detector.pb',
as_text=False)
scores = model.evaluate(test_images, test_labels, verbose=0)
print('Accuracy: %.2f%%' % scores[1] * 100)
return model
def train_and_assess_model(model, train_images, train_labels):
epochs = 60
learning_rate = 0.007
decay = learning_rate / epochs
sgd = SGD(lr=learning_rate, momentum=0.81, decay=decay, nesterov=False)
model.compile(loss='categorical_crossentropy',
optimizer=sgd,
metrics=['accuracy'])
print(model.summary())
model.fit(train_images,
train_labels,
validation_data=(train_images, train_labels),
epochs=epochs,
batch_size=12)
print(model.input.op.name)
print(model.output.op.name)
frozen_graph = freeze_graph(K.get_session(),
output_names=[model.output.op.name])
tf.train.write_graph(
frozen_graph,
'.',
'Android_App/app/src/main/assets/emotions_detector.pb',
as_text=False)
return model
def run():
happy_images = load_and_format_images('Happy')
sad_images = load_and_format_images('Sad')
train_images, train_labels, test_images, test_labels = separate_data(
happy_images, sad_images)
train_images = reshape_images(train_images)
test_images = reshape_images(test_images)
model = create_model()
model = train_and_assess_model(model, train_images, train_labels)
predict_images(model, test_images, test_labels)
plt.show(freeze_graph(K.get_session(),
output_names=[model.output.op.name]))
kernel_constraint=maxnorm(3)))
# Get the max value of each 2 x 2 block and discard the rest, thus changing image size to 16 x 16
model.add(MaxPooling2D(pool_size=(2, 2)))
# model.add(Conv2D(64, (3, 3), input_shape=(16, 16, 32), activation='relu', padding='same', kernel_constraint=maxnorm(3)))
# model.add(Conv2D(128, (3, 3), input_shape=(16, 16, 64), activation='relu', padding='same', kernel_constraint=maxnorm(3)))
# Flatten the matrix into an array
model.add(Flatten())
# Dense layer creates 512 connected neurons (32 x 16) with relu activation function
model.add(Dense(512, activation='relu', kernel_constraint=maxnorm(3)))
# Dropout layer prevents overfitting by ignoring outputs from some neurons
model.add(Dropout(0.5))
# Last dense layer sorts the features into 10 possible outputs with softmax activation
model.add(Dense(10, activation='softmax'))
# Specify loss function, optimizer, and which metrics we want to keep track of
model.compile(loss='categorical_crossentropy',
optimizer=SGD(lr=0.01),
metrics=['accuracy'])
model.fit(train_images, train_labels, epochs=30, batch_size=32)
frozen_graph = freeze_graph(K.get_session(),
output_names=[model.output.op.name])
tf.io.write_graph(frozen_graph, '.', 'cifar10.pb', as_text=False)