def augmented_train_mobilenet_with_data_generator(model_path):
"""
To train a pre-trained model.
:param model_path: pretrained model path.
:return:
"""
# STEPS_PER_EPOCH * BATCH_SIZE = DATA_SIZE
EPOCH = 5
DATA_SIZE = 100
BATCH_SIZE = 20
STEPS_PER_EPOCH = DATA_SIZE / BATCH_SIZE
CLASS_NUM = 2
INPUT_SHAPE = (224, 224)
MODEL_PATH_EACH_EPOCH = '../model/gender/gender.mobilenet.augment.{epoch:02d}-{loss:.2f}.model'
FINAL_MODEL_PATH = "../models/gender/gender.mobilenet.augment.final.model"
checkpoint = keras.callbacks.ModelCheckpoint(MODEL_PATH_EACH_EPOCH,
monitor='loss',
verbose=1)
model = keras.models.load_model(model_path,
custom_objects={
'relu6': relu6,
'DepthwiseConv2D': DepthwiseConv2D
})
X_test, Y_test = dataset.load_data_from_config(
"../data/gender/gender_test.txt", input_shape=INPUT_SHAPE)
encoder = LabelEncoder()
encoder.fit(["F", "M"])
encoded_Y_test = encoder.transform(Y_test).astype(np.int32)
ohr_Y_test = np_utils.to_categorical(encoded_Y_test, CLASS_NUM)
generator = dataset.data_generator("../data/gender/gender_train.txt",
encoder,
CLASS_NUM,
input_shape=INPUT_SHAPE,
batch_size=BATCH_SIZE)
mobilenet = MobileNet(model)
mobilenet.run_generator(generator,
loss="categorical_crossentropy",
optimizer="sgd",
epoch=EPOCH,
steps_per_epoch=STEPS_PER_EPOCH,
val_data=(X_test, ohr_Y_test),
augment=True,
callbacks=[checkpoint])
mobilenet.save(FINAL_MODEL_PATH)
def train_mobilenet_with_data_generator():
# STEPS_PER_EPOCH * BATCH_SIZE = DATA_SIZE
EPOCH = 5
DATA_SIZE = 100
BATCH_SIZE = 20
STEPS_PER_EPOCH = DATA_SIZE / BATCH_SIZE
CLASS_NUM = 2
INPUT_SHAPE = (224, 224)
MODEL_PATH_EACH_EPOCH = '../model/gender/gender.mobilenet.{epoch:02d}-{loss:.2f}.model'
FINAL_MODEL_PATH = '../models/gender/gender.mobilenet.final.model'
checkpoint = keras.callbacks.ModelCheckpoint(MODEL_PATH_EACH_EPOCH,
monitor='loss',
verbose=1)
tensorboard = keras.callbacks.TensorBoard(log_dir='../logs',
histogram_freq=0,
batch_size=8,
write_graph=True,
write_grads=False,
write_images=False,
embeddings_freq=0,
embeddings_layer_names=None,
embeddings_metadata=None)
X_test, Y_test = dataset.load_data_from_config(
"../data/gender/gender_test.txt", input_shape=INPUT_SHAPE)
encoder = LabelEncoder()
encoder.fit(["F", "M"])
encoded_Y_test = encoder.transform(Y_test).astype(np.int32)
ohr_Y_test = np_utils.to_categorical(encoded_Y_test, CLASS_NUM)
generator = dataset.data_generator("../data/gender/gender_train.txt",
encoder,
2,
input_shape=INPUT_SHAPE,
batch_size=BATCH_SIZE)
net = MobileNet()
net.build_network(include_top=True,
input_shape=(224, 224, 3),
output_shape=CLASS_NUM)
net.run_generator(generator,
loss="categorical_crossentropy",
optimizer="sgd",
epoch=EPOCH,
steps_per_epoch=STEPS_PER_EPOCH,
val_data=(X_test, ohr_Y_test),
callbacks=[checkpoint, tensorboard])
net.save(FINAL_MODEL_PATH)
def run_hybrid_xception():
genders = dataset.load_labels("../data/gender_label.txt")
births = dataset.load_labels("../data/age_label.txt")
x_test, y_test = dataset.load_data_from_config("../data/hybrid_test.txt",
input_shape=(299, 299))
gender_y_test = [y.split("_")[0] for y in y_test]
birth_y_test = [y.split("_")[1] for y in y_test]
gender_encoder = LabelEncoder()
gender_encoder.fit(genders)
encoded_gender_y_test = gender_encoder.transform(gender_y_test).astype(
np.int32)
one_hot_gender_y_test = np_utils.to_categorical(
encoded_gender_y_test,
np.max(encoded_gender_y_test) + 1)
birth_encoder = LabelEncoder()
birth_encoder.fit(births)
encoded_birth_y_test = birth_encoder.transform(birth_y_test).astype(
np.int32)
one_hot_birth_y_test = np_utils.to_categorical(
encoded_birth_y_test,
np.max(encoded_birth_y_test) + 1)
hybrid_y_test = []
for i in range(len(one_hot_gender_y_test)):
hybrid_y_test.append(
np.append(one_hot_gender_y_test[i], one_hot_birth_y_test[i]))
hybrid_y_test = np.asarray(hybrid_y_test)
# ohr_Y_train = np_utils.to_categorical(encoded_age_Y_train, np.max(encoded_age_Y_train) + 1)
# ohr_Y_test = np_utils.to_categorical(encoded_Y_test, np.max(encoded_Y_test) + 1)
network = Xception()
network.build_network(include_top=True,
input_shape=(299, 299, 3),
output_shape=hybrid_y_test[0].__len__())
network.run_generator(
dataset.data_hybrid_generator("../data/hybrid_train.txt",
gender_encoder, birth_encoder,
genders.__len__(), births.__len__(), 1),
loss="categorical_crossentropy",
optimizer="sgd",
steps_per_epoch=8,
val_data=(x_test, hybrid_y_test)) # (X_test, ohr_Y_test)
def train_mobilenet_with_data_generator(batch_size, input_shape):
class_num = birth_year.__len__()
filepath = '../models/birth/birth.mobilenet.{epoch:02d}-{loss:.2f}.model'
checkpoint = keras.callbacks.ModelCheckpoint(filepath,
monitor='loss',
verbose=1)
tensorboard = keras.callbacks.TensorBoard(log_dir='../logs',
histogram_freq=0,
batch_size=batch_size,
write_graph=True,
write_grads=False,
write_images=False,
embeddings_freq=0,
embeddings_layer_names=None,
embeddings_metadata=None)
X_test, Y_test = dataset.load_data_from_config(
"../data/birth/birth_test.txt")
encoder = LabelEncoder()
encoder.fit(birth_year)
encoded_Y_test = encoder.transform(Y_test).astype(np.int32)
ohr_Y_test = np_utils.to_categorical(encoded_Y_test, class_num)
generator = dataset.data_generator("../data/birth/birth_train.txt",
encoder,
class_num,
batch_size=batch_size)
net = MobileNet()
net.build_network(include_top=True,
input_shape=(224, 224, 3),
output_shape=class_num)
net.run_generator(generator,
loss="categorical_crossentropy",
optimizer="sgd",
epoch=1,
steps_per_epoch=10,
val_data=(X_test, ohr_Y_test),
callbacks=[checkpoint, tensorboard])
net.save("../model/birth/birth.mobilenet.h5")
def train_mobilenet_with_data_generator():
filepath = '../emotion/emotion.mobilenet.{epoch:02d}-{loss:.2f}.model'
checkpoint = keras.callbacks.ModelCheckpoint(filepath,
monitor='loss',
verbose=1)
tensorboard = keras.callbacks.TensorBoard(log_dir='../logs',
histogram_freq=0,
batch_size=8,
write_graph=True,
write_grads=False,
write_images=False,
embeddings_freq=0,
embeddings_layer_names=None,
embeddings_metadata=None)
X_test, Y_test = dataset.load_data_from_config(
"../data/emotion/emotion_test.txt", input_shape=(224, 224))
encoder = LabelEncoder()
encoder.fit(["0", "1", "2", "3", "4", "5", "6"])
encoded_Y_test = encoder.transform(Y_test).astype(np.int32)
ohr_Y_test = np_utils.to_categorical(encoded_Y_test, 7)
generator = dataset.data_generator("../data/emotion/emotion_train.txt",
encoder,
7,
input_shape=(224, 224),
batch_size=20)
net = MobileNet()
net.build_network(include_top=True,
input_shape=(224, 224, 3),
output_shape=7)
net.run_generator(generator,
loss="categorical_crossentropy",
optimizer="sgd",
epoch=2,
steps_per_epoch=10,
val_data=(X_test, ohr_Y_test),
callbacks=[checkpoint, tensorboard])
net.save("../models/emotion/emotion.mobilenet.final.model")