def build_egomotion_test_graph(self):
"""Builds egomotion model reading from placeholders."""
input_image_stack = tf.placeholder(
tf.float32,
[1, self.img_height, self.img_width, self.seq_length * 3],
name='raw_input')
input_bottleneck_stack = None
if self.imagenet_norm:
im_mean = tf.tile(
tf.constant(reader.IMAGENET_MEAN), multiples=[self.seq_length])
im_sd = tf.tile(
tf.constant(reader.IMAGENET_SD), multiples=[self.seq_length])
input_image_stack = (input_image_stack - im_mean) / im_sd
if self.joint_encoder:
# Pre-compute embeddings here.
with tf.variable_scope('depth_prediction', reuse=True):
input_bottleneck_stack = []
encoder_selected = nets.encoder(self.architecture)
for i in range(self.seq_length):
input_image = input_image_stack[:, :, :, i * 3:(i + 1) * 3]
tf.get_variable_scope().reuse_variables()
embedding, _ = encoder_selected(
target_image=input_image,
weight_reg=self.weight_reg,
is_training=True)
input_bottleneck_stack.append(embedding)
input_bottleneck_stack = tf.concat(input_bottleneck_stack, axis=3)
with tf.variable_scope('egomotion_prediction'):
est_egomotion = nets.egomotion_net(
image_stack=input_image_stack,
disp_bottleneck_stack=input_bottleneck_stack,
joint_encoder=self.joint_encoder,
seq_length=self.seq_length,
weight_reg=self.weight_reg,
same_trans_rot_scaling=self.same_trans_rot_scaling)
self.input_image_stack = input_image_stack
self.est_egomotion = est_egomotion
def build_egomotion_test_graph(self):
"""Builds egomotion model reading from placeholders."""
input_image_stack = tf.placeholder(
tf.float32,
[1, self.img_height, self.img_width, self.seq_length * 3],
name='raw_input')
input_bottleneck_stack = None
if self.imagenet_norm:
im_mean = tf.tile(
tf.constant(reader.IMAGENET_MEAN), multiples=[self.seq_length])
im_sd = tf.tile(
tf.constant(reader.IMAGENET_SD), multiples=[self.seq_length])
input_image_stack = (input_image_stack - im_mean) / im_sd
if self.joint_encoder:
# Pre-compute embeddings here.
with tf.variable_scope('depth_prediction', reuse=True):
input_bottleneck_stack = []
encoder_selected = nets.encoder(self.architecture)
for i in range(self.seq_length):
input_image = input_image_stack[:, :, :, i * 3:(i + 1) * 3]
tf.get_variable_scope().reuse_variables()
embedding, _ = encoder_selected(
target_image=input_image,
weight_reg=self.weight_reg,
is_training=True)
input_bottleneck_stack.append(embedding)
input_bottleneck_stack = tf.concat(input_bottleneck_stack, axis=3)
with tf.variable_scope('egomotion_prediction'):
est_egomotion = nets.egomotion_net(
image_stack=input_image_stack,
disp_bottleneck_stack=input_bottleneck_stack,
joint_encoder=self.joint_encoder,
seq_length=self.seq_length,
weight_reg=self.weight_reg)
self.input_image_stack = input_image_stack
self.est_egomotion = est_egomotion
def build_egomotion_test_graph(self):
"""Builds egomotion model reading from placeholders."""
print('EGOMOTION PROCESS')
input_image_stack = tf.placeholder(
tf.float32,
[1, self.img_height, self.img_width, self.seq_length * 3],
name='raw_input')
input_bottleneck_stack = None
if self.imagenet_norm:
im_mean = tf.tile(
tf.constant(reader.IMAGENET_MEAN), multiples=[self.seq_length])
im_sd = tf.tile(
tf.constant(reader.IMAGENET_SD), multiples=[self.seq_length])
input_image_stack = (input_image_stack - im_mean) / im_sd
if self.joint_encoder:
# Pre-compute embeddings here.
with tf.variable_scope('depth_prediction', reuse=True):
input_bottleneck_stack = []
encoder_selected = nets.encoder(self.architecture)
for i in range(self.seq_length):
input_image = input_image_stack[:, :, :, i * 3:(i + 1) * 3]
tf.get_variable_scope().reuse_variables()
embedding, _ = encoder_selected(
target_image=input_image,
weight_reg=self.weight_reg,
is_training=True)
input_bottleneck_stack.append(embedding)
input_bottleneck_stack = tf.concat(input_bottleneck_stack, axis=3)
with tf.variable_scope('egomotion_prediction'):
est_egomotion = nets.egomotion_net(
image_stack=input_image_stack,
disp_bottleneck_stack=input_bottleneck_stack,
joint_encoder=self.joint_encoder,
seq_length=self.seq_length,
weight_reg=self.weight_reg)
# sess = tf.Session()
# with sess.as_default():
# tensor_new = sess.run(est_egomotion)
# print(tensor_new)
# with tf.Session() as sess:
# init = tf.global_variables_initializer()
# sess.run(init)
# print(est_egomotion.eval())
print('oke oce')
print(est_egomotion.get_shape()) # Call the shape of the tensor
# x = tf.Print(est_egomotion, [est_egomotion])
# sess = tf.InteractiveSession()
# sess.run(x)
# aselole = tf.Print(est_egomotion, [est_egomotion], "ehehhe")
# sess = tf.Session()
# print(sess.run(aselole))
print('Flag 1')
print('est_egomotionssss = ', est_egomotion)
print('Flag 2')
print(est_egomotion)
# sess = tf.Session()
# with sess.as_default():
# tensor_new = sess.run(egomotion_prediction)
# print(tensor_new)
self.input_image_stack = input_image_stack
self.est_egomotion = est_egomotion
print('est_egomotionssss = ', self.est_egomotion)