def _parse_ucf_features_test(record):
features = {
"img": tf.FixedLenFeature((), tf.string, default_value=''),
"label": tf.FixedLenFeature((), tf.float32, default_value=0),
"score": tf.FixedLenFeature((), tf.int64, default_value=0),
"width": tf.FixedLenFeature((), tf.int64, default_value=0),
"height": tf.FixedLenFeature((), tf.int64, default_value=0),
"channel": tf.FixedLenFeature((), tf.int64, default_value=0)
}
parsed_features = tf.parse_single_example(record, features)
# for key in parsed_features:
# print(key, type(parsed_features[key]))
# print(type(parsed_features['img']))
img = tf.decode_raw(parsed_features['img'], tf.float32)
img_reshape = tf.reshape(img, (tf.stack([
parsed_features['width'], parsed_features['height'],
parsed_features['channel']
])))
img_reshape = inception_preprocessing.preprocess_for_eval(
img_reshape,
image_size,
image_size,
central_fraction=1,
scope='preprocess_test')
return img_reshape, parsed_features['width'], parsed_features['height'], parsed_features['channel'], \
parsed_features['score'] * 5, parsed_features['label'] * 20
def _parse_ucf_features_train(record):
features = {"img": tf.FixedLenFeature((), tf.string, default_value=''),
"label": tf.FixedLenFeature((), tf.float32, default_value=0),
"score": tf.FixedLenFeature((), tf.int64, default_value=0),
"width": tf.FixedLenFeature((), tf.int64, default_value=0),
"height": tf.FixedLenFeature((), tf.int64, default_value=0),
"channel": tf.FixedLenFeature((), tf.int64, default_value=0)}
parsed_features = tf.parse_single_example(record, features)
img = tf.decode_raw(parsed_features['img'], tf.float32)
img_reshape = tf.reshape(img, (
tf.stack([parsed_features['width'], parsed_features['height'], parsed_features['channel']])))
img_reshape = tf.image.random_flip_left_right(img_reshape)
img_reshape = tf.image.random_brightness(img_reshape, max_delta=10. / 255.)
img_reshape = tf.image.random_saturation(img_reshape, lower=0.9, upper=1.1)
angles = tf.random_uniform(shape=(1,), minval=-0.2, maxval=0.2,
dtype=tf.float32,
seed=tf.set_random_seed(1234), name=None)
noise = tf.random_normal(shape=tf.shape(img_reshape), mean=0.0, stddev=0.01, dtype=tf.float32)
img_reshape = img_reshape + noise
img_reshape = tf.contrib.image.rotate(tf.expand_dims(img_reshape, 0), angles)
img_reshape = inception_preprocessing.preprocess_for_eval(
tf.squeeze(img_reshape), image_size, image_size,
central_fraction=0.9, scope='preprocess_train')
return img_reshape, parsed_features['width'], parsed_features['height'], parsed_features['channel'], \
parsed_features['score'] * 5, parsed_features['label'] * 20
def preprocess_image(image_buffer,
output_height,
output_width,
num_channels=_NUM_CHANNELS,
is_training=False):
if is_training:
return preprocess_for_train(image_buffer, output_height, output_width,
num_channels)
else:
return preprocess_for_eval(image_buffer, output_height, output_width,
num_channels)
def build_model(graph):
with graph.as_default():
with tf.contrib.slim.arg_scope(resnet.inception_resnet_v2_arg_scope()):
placeholder = tf.placeholder(tf.string, name='input')
image = tf.image.decode_jpeg(placeholder, channels=3, name='image')
image = preprocess.preprocess_for_eval(image, FLAGS.image_size,
FLAGS.image_size)
image = tf.expand_dims(image, 0)
logits, end_points = resnet.inception_resnet_v2(
image, is_training=False, dropout_keep_prob=1.0)
saver = tf.train.Saver(tf.global_variables())
return placeholder, logits, end_points
def getImage(self, expandDim):
if( self.inputH is None ):
raise ValueError("input height not set")
if( self.inputC is not 3):
raise ValueError("Only ch=3 implemented so far")
if( self.imageUrl is None):
raise ValueError("Image URL is not set")
if( self.inputH != self.inputW):
raise ValueError("Non-square inputs not yet implemented")
try:
f = urllib.request.urlopen(self.imageUrl)
jpeg_str = f.read()
image = Image.open(BytesIO(jpeg_str))
except IOError:
raise ValueError('Cannot retrieve image. Please check url: ' + self.imageUrl)
image = np.asarray(image.convert('RGB'))
print("image shape: ", image.shape)
m = min(image.shape[0], image.shape[1])
print("min: ", m)
if(self.preprocessingType == "vgg"):
image = vgg_preprocessing.preprocess_for_eval(image, self.inputH, self.inputW, m)
elif(self.preprocessingType == "inception"):
# image = image.astype("float32")
# with tf.Session() as sess:
# image = image.eval(session=sess)
image = tf.convert_to_tensor(image)
image = inception_preprocessing.preprocess_for_eval(image, height=self.inputH, width=self.inputW)
elif(self.preprocessingType == "resize_only"):
image = np.expand_dims(image, 0)
image = tf.image.resize_bilinear(image, [self.inputH, self.inputW], align_corners=False)
image = tf.squeeze(image, axis=0)
# image = tf.convert_to_tensor(image)
else:
raise ValueError("Unknown preprocessing type: ", self.preprocessingType)
# image = tf.expand_dims(image, 0)
print("new shape: ", image.shape)
print("new type: ", type(image))
with tf.Session() as sess:
image = image.eval(session=sess)
# print("new shape2: ", image.shape)
# print("new type2: ", type(image))
if(len(image.shape) == 3 and expandDim):
image = np.expand_dims(image, 0)
print("new shape3: ", image.shape)
print("new type3: ", type(image))
return image
def __init__(self, weights_file):
config = tf.ConfigProto(log_device_placement=False)
config.gpu_options.allow_growth = True
self.__graph = tf.Graph()
self.image_size = inception_v4.inception_v4.default_image_size
with self.__graph.as_default():
self.__session = tf.Session(config=config, graph=self.__graph)
self.images_placeholder = tf.placeholder(
tf.float32,
shape=(None, self.image_size, self.image_size, 3),
name='image')
self.phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
arg_scope = inception_v4.inception_v4_arg_scope()
with slim.arg_scope(arg_scope):
logits, end_points = inception_v4.inception_v4(
self.images_placeholder,
is_training=False,
num_classes=1001,
create_aux_logits=False)
for tensor in tf.all_variables():
print(tensor.name, tensor.shape)
self.predict = tf.nn.softmax(logits)
self.end_points = end_points
# preprocess
with tf.device('/cpu:0'):
self.one_image_placeholder = tf.placeholder(tf.float32,
shape=(None, None,
3),
name='one_image')
self.preprocessed_image = inception_preprocessing.preprocess_for_eval(
self.one_image_placeholder,
self.image_size,
self.image_size,
central_fraction=1)
print('[{}]:InceptionV4 model restoring...'.format(
datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")))
saver = tf.train.Saver(tf.global_variables(), max_to_keep=3)
saver.restore(self.__session, weights_file)
self.__session.graph.finalize()
print('[{}]:InceptionV4 model restore succeed!'.format(
datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")))
def __init__(self, weight_file):
config = tf.ConfigProto(log_device_placement=False)
config.gpu_options.allow_growth = True
self.__graph = tf.Graph()
self.image_size = inception_v4.inception_v4.default_image_size
with self.__graph.as_default():
self.__session = tf.Session(config=config, graph=self.__graph)
self.images_placeholder = tf.placeholder(
tf.float32,
shape=(None, self.image_size, self.image_size, 3),
name='image')
self.phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
arg_scope = inception_v4.inception_v4_arg_scope()
with slim.arg_scope(arg_scope):
logits, end_points = inception_v4.inception_v4(
self.images_placeholder,
is_training=False,
num_classes=1001)
for tensor in tf.all_variables():
print(tensor.name, tensor.shape)
self.predict = tf.nn.softmax(logits)
self.end_points = end_points
# preprocess
self.one_image_placeholder = tf.placeholder(tf.float32,
shape=(None, None, 3),
name='one_image')
self.preprocessed_image = inception_preprocessing.preprocess_for_eval(
self.one_image_placeholder,
self.image_size,
self.image_size,
central_fraction=1)
saver = tf.train.Saver(tf.global_variables(), max_to_keep=3)
saver.restore(self.__session, weight_file)
def preprocess_frame(frame_path, gif_path, train_mode):
print("function: preprocess_frame()")
if train_mode == True:
gif_path = os.path.join(train_path, gif_path)
else:
gif_path = os.path.join(test_path, gif_path)
frame_path = os.path.join(gif_path, frame_path)
frame_contents = tf.read_file(frame_path)
print(frame_contents)
frame_tensor = tf.image.decode_png(frame_contents)
print(frame_tensor)
# print input tensor shape??
print("frame tensor shape:", frame_tensor.get_shape())
# print tensor everytime it's modified
# resized_frame = tf.image.resize_images(frame_tensor,299,299)
frame_tensor = tf.reshape(frame_tensor, (299, 299, 3))
# frame_tensor = 2*(frame_tensor/tf.cast(255.0,tf.uint8))-tf.cast(1.0,tf.uint8)
frame_tensor = preprocess_for_eval(frame_tensor, height=227, width=227)
# frame_tensor = tf.cast(frame_tensor, uint8)
return frame_tensor
def load_model():
tf.reset_default_graph()
slim = tf.contrib.slim
height = 299
width = 299
channels = 3
X = tf.placeholder(shape=[None, None, None, 3], dtype=tf.float32)
X_processed = tf.map_fn(
lambda x: pre.preprocess_for_eval(image=x, width=width, height=height),
X)
with slim.arg_scope(inception_resnet_v2_arg_scope()):
inception_resnet_v2(X_processed, num_classes=1001, is_training=False)
saver = tf.train.Saver()
sess = tf.Session()
saver.restore(sess, "inception_resnet_v2_2016_08_30.ckpt")
graph = sess.graph
mid_layer = graph.get_tensor_by_name(
"InceptionResnetV2/AuxLogits/Conv2d_2a_5x5/BatchNorm/FusedBatchNorm:0")
mid_layer_a = tf.contrib.layers.flatten(mid_layer)
return (sess, X, mid_layer_a)
def __init__(self, weight_file):
self.image_size = inception_v4.inception_v4.default_image_size
config = tf.ConfigProto(log_device_placement=False)
config.gpu_options.allow_growth = True
self.__graph = tf.Graph()
with self.__graph.as_default():
self.__session = tf.Session(config=config, graph=self.__graph)
self.images_placeholder = tf.placeholder(tf.float32,
shape=(None, None, 3),
name='image')
img_process = inception_preprocessing.preprocess_for_eval(
self.images_placeholder,
self.image_size,
self.image_size,
central_fraction=1,
scope='preprocess_test')
img_process = tf.expand_dims(img_process, 0)
arg_scope = inception_v4.inception_v4_arg_scope()
with tf.variable_scope('big', reuse=False) as scope:
with slim.arg_scope(arg_scope):
logits, end_points = inception_v4.inception_v4(
img_process,
is_training=False,
num_classes=1001,
dropout_keep_prob=1.0,
reuse=False,
create_aux_logits=False)
with tf.variable_scope('Beauty', 'BeautyV1',
reuse=False) as scope:
mid_full_conn = slim.fully_connected(
end_points['PreLogitsFlatten'],
1000,
activation_fn=tf.nn.relu,
scope='mid_full_conn',
trainable=False,
reuse=False)
predict_conn = slim.fully_connected(mid_full_conn,
100,
activation_fn=None,
scope='100_class_conn',
trainable=False,
reuse=False)
beauty_weight = tf.convert_to_tensor(
[[i] for i in range(0, 100)], dtype=tf.float32)
regress_conn = tf.matmul(tf.nn.softmax(predict_conn),
beauty_weight) # 32 * 1
self.end_points = {
'mid_full_conn': mid_full_conn,
'regress_conn': regress_conn,
'predict_conn': predict_conn
}
saver = tf.train.Saver(tf.global_variables(), max_to_keep=3)
saver.restore(self.__session, weight_file)