def getData(self):
if self.config['dataset'] == 'voc':
self.data_model = VOC()
self.data_model.getFileList()
self.data_model.getAnnot()
self.data_model.getClassAnnot()
self.data_model.setGeneratorConfig(self.config)
def Xml2Txt():
voc_dataset = VOC()
with open(file=Config.txt_file_dir, mode="a+", encoding="utf-8") as f:
for i, sample in enumerate(voc_dataset):
num_bboxes = len(sample["bboxes"])
line_text = sample["image_file_dir"] + " " + str(
sample["image_height"]) + " " + str(
sample["image_width"]) + " "
for j in range(num_bboxes):
bbox = list(map(str, sample["bboxes"][j]))
cls = str(sample["class_ids"][j])
bbox.append(cls)
line_text += " ".join(bbox)
line_text += " "
line_text = line_text.strip()
line_text += "\n"
print("Writing information of picture {} to {}".format(
sample["image_file_dir"], Config.txt_file_dir))
f.write(line_text)
from data.voc import VOC
from configuration import Config
if __name__ == '__main__':
voc_dataset = VOC()
with open(file=Config.txt_file_dir, mode="a+", encoding="utf-8") as f:
for i, sample in enumerate(voc_dataset):
num_bboxes = len(sample["bboxes"])
line_text = sample["image_file_dir"] + " " + str(
sample["image_height"]) + " " + str(
sample["image_width"]) + " "
for j in range(num_bboxes):
bbox = list(map(str, sample["bboxes"][j]))
cls = str(sample["class_ids"][j])
bbox.append(cls)
line_text += " ".join(bbox)
line_text += " "
line_text = line_text.strip()
line_text += "\n"
print("Writing information of picture {} to {}".format(
sample["image_file_dir"], Config.txt_file_dir))
f.write(line_text)
def getData(self):
self.voc = VOC(VOC.VOC_ALL)
self.voc.getFileList()
self.voc.getAnnot()
self.voc.getClassAnnot()
self.voc.setGeneratorConfig(self.config)
class Alexnet_Train(Train):
def focal_loss(gamma=2., alpha=.25):
def focal_loss_fixed(y_true, y_pred):
pt_1 = tf.where(tf.equal(y_true, 1), y_pred, tf.ones_like(y_pred))
pt_0 = tf.where(tf.equal(y_true, 0), y_pred, tf.zeros_like(y_pred))
return -K.sum(
alpha * K.pow(1. - pt_1, gamma) * K.log(pt_1)) - K.sum(
(1 - alpha) * K.pow(pt_0, gamma) * K.log(1. - pt_0))
return focal_loss_fixed
def __init__(self):
self.config = config
def getData(self):
self.voc = VOC(VOC.VOC_ALL)
self.voc.getFileList()
self.voc.getAnnot()
self.voc.getClassAnnot()
self.voc.setGeneratorConfig(self.config)
def initModel(self):
config = tf.ConfigProto(device_count={'GPU': 1})
sess = tf.Session(config=config)
K.set_session(sess)
self.model = Alexnet.set(include_inputs=True,
class_num=NUM_CLASSES,
input_dim=INPUT_DIM,
output=self.config['output_func'])
self.model.summary()
def buildTrainKeras(self):
#optimizer = Adam(lr=1e-4, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
optimizer = SGD(lr=self.config['lr'],
decay=self.config['decay'],
momentum=0.9)
#optimizer = RMSprop(lr=1e-5, rho=0.9, epsilon=1e-08, decay=0.0)
self.model.compile(
loss=self.config['loss'], #'binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
def fit(self):
file = path_name + '/' + model_name + '-best.h5'
if os.path.exists(file):
self.model.load_weights(file)
early_stop = EarlyStopping(monitor='val_loss',
min_delta=0.0001,
patience=20,
mode='min',
verbose=1)
checkpoint = ModelCheckpoint(path_name + '/' + model_name + '-best.h5',
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
tb_counter = 1
tensorboard = TensorBoard(log_dir=log_path + '/' + model_name + '_' +
str(tb_counter),
histogram_freq=0,
write_graph=True,
write_images=False)
reduce_lr = ReduceLROnPlateau(monitor='val_loss',
factor=0.2,
patience=2,
min_lr=0.000001,
verbose=1,
cooldown=1)
train_batch = self.voc.getTrainBatch()
valid_batch = self.voc.getValidBatch()
if self.config['data_mode'] == 'on_memory':
tr_x, tr_y = train_batch
tst_x, tst_y = valid_batch
self.history = self.model.fit(
tr_x,
tr_y,
batch_size=self.config['batch_size'],
epochs=self.config['epochs'],
verbose=1,
validation_data=(tst_x, tst_y),
callbacks=[early_stop, checkpoint, tensorboard, reduce_lr],
max_queue_size=3)
else:
self.history = self.model.fit_generator(
generator=train_batch,
steps_per_epoch=len(train_batch),
epochs=self.config['epochs'],
verbose=1,
validation_data=valid_batch,
validation_steps=len(valid_batch),
callbacks=[early_stop, checkpoint, tensorboard, reduce_lr],
max_queue_size=3)
def run(self):
self.initModel()
self.getData()
self.buildTrainKeras()
self.fit()
self.save(file_name=file_name, path_name=path_name)
class Mobilenet_Train(Train):
def __init__(self):
self.config = config
def getData(self):
if self.config['dataset'] == 'voc':
self.data_model = VOC()
self.data_model.getFileList()
self.data_model.getAnnot()
self.data_model.getClassAnnot()
self.data_model.setGeneratorConfig(self.config)
def initModel(self):
_config = tf.ConfigProto(device_count={'GPU': 1})
sess = tf.Session(config=_config)
K.set_session(sess)
if self.config['model_info'] == 'begin':
self.model = Mobilenetv2.set(include_inputs=True,
class_num=NUM_CLASSES,
input_dim=INPUT_DIM,
output='sigmoid')
elif self.config['model_info'] == 'pretained':
self.model = Mobilenetv2.getKerasModelBase(num_class=NUM_CLASSES,
output='sigmoid',
fix_layer=156)
self.model.summary()
def buildTrainKeras(self):
optimizer = Adam(lr=1e-4,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08,
decay=0.0)
#optimizer = SGD(lr=1e-4, decay=0.0005, momentum=0.9)
#optimizer = RMSprop(lr=1e-5, rho=0.9, epsilon=1e-08, decay=0.0)
self.model.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
def fit(self):
file = path_name + '/' + model_name + '-best.h5'
if os.path.exists(file):
self.model.load_weights(file)
fix_layer = 156
for layer in self.model.layers:
layer.trainable = False
# or if we want to set the first 20 layers of the network to be non-trainable
for layer in self.model.layers[:fix_layer]:
layer.trainable = False
for layer in self.model.layers[fix_layer:]:
layer.trainable = True
early_stop = EarlyStopping(monitor='val_loss',
min_delta=0.001,
patience=3,
mode='min',
verbose=1)
checkpoint = ModelCheckpoint(file,
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
tb_counter = 1
tensorboard = TensorBoard(log_dir=log_path + '/' + model_name + '_' +
str(tb_counter),
histogram_freq=0,
write_graph=True,
write_images=False)
train_batch = self.data_model.getTrainBatch()
valid_batch = self.data_model.getValidBatch()
self.model.fit_generator(
generator=train_batch,
steps_per_epoch=len(train_batch),
epochs=self.config['epochs'],
verbose=1,
validation_data=valid_batch,
validation_steps=len(valid_batch),
callbacks=[early_stop, checkpoint, tensorboard],
max_queue_size=3)
def run(self):
self.initModel()
self.getData()
self.buildTrainKeras()
self.fit()
self.save(file_name=file_name, path_name=path_name)