def __init__(self):
# im, all_coordinates = demo.detection_demo(im_names)
global sess, net, keras_model_antnet,keras_model_back,keras_model_face,\
classes_cows_back,classes_cows_antnet,classes_cows_face, graph_face, graph_back,graph_antnet,graph_objdetect
sess, net = demo.restore_model('60000')
graph_objdetect = tf.get_default_graph()
# tf.reset_default_graph()
# sess_id, net_id = demo_id.restore_model('100000')
# set the backend whether tf or theano
util.set_img_format()
config.model = 'xception_face'
# get the keras original model object
keras_module_face = util.get_model_class_instance()
# load the best trained model
keras_model_face = keras_module_face.load()
graph_face = tf.get_default_graph()
# get the classes dict
classes_cows_face = util.get_classes_in_keras_format()
config.model = 'xception_back'
# get the keras original model object
keras_module_back = util.get_model_class_instance()
# load the best trained model
keras_model_back = keras_module_back.load()
graph_back = tf.get_default_graph()
# get the classes dict
classes_cows_back = util.get_classes_in_keras_format()
config.model = 'antnet'
# get the keras original model object
keras_module_antnet = util.get_model_class_instance()
# load the best trained model
keras_model_antnet = keras_module_antnet.load()
graph_antnet = tf.get_default_graph()
# get the classes dict
classes_cows_antnet = util.get_classes_in_keras_format()
# warm up the model
print('Warming up the keras model face')
input_shape = (1, ) + keras_module_face.img_size + (3, )
dummpy_img = numpy.ones(input_shape)
dummpy_img = preprocess_input(dummpy_img)
keras_model_face.predict(dummpy_img)
# warm up the model
print('Warming up the keras model back')
input_shape = (1, ) + keras_module_back.img_size + (3, )
dummpy_img = numpy.ones(input_shape)
dummpy_img = preprocess_input(dummpy_img)
keras_model_back.predict(dummpy_img)
# warm up the model
print('Warming up the keras model antnet')
input_shape = (1, ) + keras_module_antnet.img_size + (3, )
dummpy_img = numpy.ones(input_shape)
dummpy_img = preprocess_input(dummpy_img)
keras_model_antnet.predict([dummpy_img, np.random.rand(1)])
def train_nn():
df = pd.read_csv(config.activations_path)
df, y, classes = encode(df)
X_train, X_test, y_train, y_test = train_test_split(df.values, y, test_size=0.2, random_state=17)
model_module = util.get_model_class_instance()
model = Sequential()
model.add(Dense(48, input_dim=model_module.noveltyDetectionLayerSize, activation='elu', init='uniform'))
model.add(Dropout(0.5))
model.add(Dense(32, activation='elu', init='uniform'))
model.add(Dropout(0.5))
model.add(Dense(len(classes), activation='softmax', init='uniform'))
model.compile(loss='sparse_categorical_crossentropy', optimizer='sgd', metrics=['accuracy'])
early_stopping = EarlyStopping(verbose=1, patience=40, monitor='val_loss')
model_checkpoint = ModelCheckpoint(config.get_novelty_detection_model_path(), save_best_only=True,
save_weights_only=True, monitor='val_loss')
callbacks_list = [early_stopping, model_checkpoint]
model.fit(
X_train,
y_train,
nb_epoch=300,
validation_data=(X_test, y_test),
batch_size=16,
callbacks=callbacks_list)
out = model.predict(X_test)
predictions = np.argmax(out, axis=1)
print("Accuracy: ", accuracy_score(y_test, predictions))
def train(nb_epoch, freeze_layers_number):
model = util.get_model_class_instance(
class_weight=util.get_class_weight(config.train_dir),
nb_epoch=nb_epoch,
freeze_layers_number=freeze_layers_number)
model.train()
print('Training is finished!')
def train_nn():
df = pd.read_csv(config.activations_path)
df, y, classes = encode(df)
X_train, X_test, y_train, y_test = train_test_split(df.values, y, test_size=0.2, random_state=17)
model_module = util.get_model_class_instance()
model = Sequential()
model.add(Dense(48, input_dim=model_module.noveltyDetectionLayerSize, activation='elu', init='uniform'))
model.add(Dropout(0.5))
model.add(Dense(32, activation='elu', init='uniform'))
model.add(Dropout(0.5))
model.add(Dense(len(classes), activation='softmax', init='uniform'))
model.compile(loss='sparse_categorical_crossentropy', optimizer='sgd', metrics=['accuracy'])
early_stopping = EarlyStopping(verbose=1, patience=40, monitor='val_loss')
model_checkpoint = ModelCheckpoint(config.get_novelty_detection_model_path(), save_best_only=True,
save_weights_only=True, monitor='val_loss')
callbacks_list = [early_stopping, model_checkpoint]
model.fit(
X_train,
y_train,
nb_epoch=300,
validation_data=(X_test, y_test),
batch_size=1,
callbacks=callbacks_list)
out = model.predict(X_test)
predictions = np.argmax(out, axis=1)
print("Accuracy: ", accuracy_score(y_test, predictions))
def train():
model = util.get_model_class_instance(
class_weight=util.get_class_weight(config.train_dir),
nb_epoch=args.nb_epoch,
batch_size=args.batch_size,
freeze_layers_number=args.freeze_layers_number)
model.train()
print('Training is finished!')
def train():
model = util.get_model_class_instance(
class_weight=util.get_class_weight(config.train_dir),
nb_epoch=args.nb_epoch,
batch_size=args.batch_size,
freeze_layers_number=args.freeze_layers_number)
model.train()
print('Training is finished!')
def train(nb_epoch,
freeze_layers_number,
auto_load_finetune=False,
visual=False):
model = util.get_model_class_instance(
class_weight=util.get_class_weight(config.train_dir),
nb_epoch=nb_epoch,
freeze_layers_number=freeze_layers_number)
model.train(auto_load_fine_tune=auto_load_finetune, visual=visual)
print('Training is finished!')
def predict_target_by_paths(paths, augment_times=1):
util.set_img_format()
# sys.argv[0] = "--model=resnet50"
predict.args = predict.parse_args()
predict.model_module = util.get_model_class_instance()
predict.model = predict.model_module.load()
predict.classes_in_keras_format = util.get_classes_in_keras_format()
for path in paths:
predict.predict(dir=path, augment_times=augment_times)
def run():
model = util.get_model_class_instance(
class_weight=util.get_class_weight(config.train_dir),
nb_epoch=args.nb_epoch,
batch_size=args.batch_size,
freeze_layers_number=args.freeze_layers_number)
#config.train_dir = 'data/sorted/test/'
#model.load()
#model.test()
model.train()
#model.extract()
model.evaluate()
print('Training is finished!')
def predict_target_by_dir(main_dir, augment_times=1):
util.set_img_format()
# sys.argv[0] = "--model=resnet50"
predict.args = predict.parse_args()
predict.model_module = util.get_model_class_instance()
predict.model = predict.model_module.load()
predict.classes_in_keras_format = util.get_classes_in_keras_format()
dirs = glob.glob(main_dir + "*")
global_summary = {"total": 0, "trues": 0, "falses": 0, "acc": 0}
iter = 0
for dir in dirs:
stats = predict.predict(dir + os.sep, iter, augment_times, False)
iter += 1
global_summary["total"] += stats["summary"]["total"]
global_summary["trues"] += stats["summary"]["trues"]
global_summary["falses"] += stats["summary"]["falses"]
global_summary["acc"] = float(
global_summary["trues"]) / global_summary["total"]
print("Global summary: {0}".format(global_summary))
if __name__ == '__main__':
tic = time.clock()
args = parse_args()
print('=' * 50)
print('Called with args:')
print(args)
if args.data_dir:
config.data_dir = args.data_dir
config.set_paths()
if args.model:
config.model = args.model
util.set_img_format()
model_module = util.get_model_class_instance()
model = model_module.load()
classes_in_keras_format = util.get_classes_in_keras_format()
all_metrix = []
print('args.path:', os.listdir(args.path))
for cow_dir in os.listdir(args.path):
root = args.path + str(cow_dir)
store_label = predict(root)
# print(store_label)
all_metrix.append(metrix(store_label))
total = 0
for i in range(len(all_metrix)):
total += all_metrix[i]
try:
args = parse_args()
if args.data_dir:
config.data_dir = args.data_dir
config.set_paths()
if args.model:
config.model = args.model
util.lock()
util.override_keras_directory_iterator_next()
util.set_classes_from_train_dir()
util.set_samples_info()
if not os.path.exists(config.trained_dir):
os.mkdir(config.trained_dir)
class_weight = util.get_class_weight(config.train_dir)
# TODO: create class instance without dynamic module import
model = util.get_model_class_instance(
class_weight=class_weight,
nb_epoch=args.nb_epoch,
freeze_layers_number=args.freeze_layers_number)
model.train()
print('Training is finished!')
except (KeyboardInterrupt, SystemExit):
util.unlock()
except Exception as e:
print(e)
print(traceback.format_exc())
util.unlock()
normalize=True)
if __name__ == '__main__':
tic = time.clock() #return the current cpu time
args = parse_args()
print('=' * 50)
print('Called with args:')
print(args)
if args.data_dir: #user defined directory
config.data_dir = args.data_dir #~
config.set_paths() #~
if args.model: #user defined model
config.model = args.model #~
util.set_img_format() #channel_first or channels_last
model_module = util.get_model_class_instance() #class model.resnet50
model = model_module.load(
) #creat base_model and load trained weight!(ResNet50) "G:\keras-transfer-learning-for-oxford102\trained\fine-tuned-resnet50-weights.h5"
classes_in_keras_format = util.get_classes_in_keras_format(
) #get a dictory of classes
predict(args.path) #we must input a path of directory including pictures
if args.execution_time: #
toc = time.clock() #record current time
print('Time: %s' % (toc - tic)) #print execution time
cnf_matrix = confusion_matrix(y_trues, predictions)
util.plot_confusion_matrix(cnf_matrix, config.classes, normalize=False)
util.plot_confusion_matrix(cnf_matrix, config.classes, normalize=True)
if __name__ == '__main__':
tic = time.clock()
args = parse_args()
print('=' * 50)
print('Called with args:')
print(args)
if args.data_dir:
config.data_dir = args.data_dir
config.set_paths()
if args.model:
config.model = args.model
util.set_img_format()
model_module = util.get_model_class_instance()
model = model_module.load()
classes_in_keras_format = util.get_classes_in_keras_format()
predict(args.path)
if args.execution_time:
toc = time.clock()
print('Time: %s' % (toc - tic))
print("Test size: ", len(test_data))
v = (len(test_data) // batch_size)
print("Loaded images paths and labels!")
try:
args = parse_args()
if args.data_dir:
config.data_dir = args.data_dir
config.set_paths()
if args.model:
config.model = args.model
init()
model = util.get_model_class_instance(
class_weight=util.get_class_weight(config.train_dir),
nb_epoch=args.nb_epoch,
batch_size=args.batch_size,
freeze_layers_number=args.freeze_layers_number)
model.load()
print("Loaded Model! Now compiling...")
model.model.compile(loss='categorical_crossentropy',
optimizer=Adam(lr=1e-5),
metrics=['accuracy'])
model.model.summary()
print("Compiled! now evaluating")
model.model.evaluate_generator(imageLoader(test_data, batch_size),
steps=v)
except Exception as e:
print(e)
traceback.print_exc()
