Skip to content

forward_gradient in get_loss should not be trained #3

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
AmazingAng wants to merge 1 commit into
base: master
Choose a base branch
from
Open

forward_gradient in get_loss should not be trained #3

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 29 additions & 2 deletions tvnet.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,7 +125,34 @@ def forward_gradient(self, x, name):
diff_y = tf.concat(axis=1, values=[diff_y_valid, last_row])

return diff_x, diff_y

def forward_gradient_forloss(self, x, name):
assert len(x.shape) == 4

with tf.variable_scope('forward_gradient'):
x_ker_init = tf.constant_initializer([[-1, 1]])
diff_x = tf.layers.conv2d(x, x.shape[-1].value, [1, 2], padding='same',
kernel_initializer=x_ker_init, use_bias=False, name=name + '_diff_x',
trainable=True)

y_ker_init = tf.constant_initializer([[-1], [1]])
diff_y = tf.layers.conv2d(x, x.shape[-1].value, [2, 1], padding='same',
kernel_initializer=y_ker_init, use_bias=False, name=name + '_diff_y',
trainable=True)

# refine the boundary
diff_x_valid = tf.slice(diff_x, begin=[0, 0, 0, 0],
size=[-1, x.shape[1].value, x.shape[2].value - 1, x.shape[3].value])
last_col = tf.zeros([tf.shape(x)[0], x.shape[1].value, 1, x.shape[3].value], dtype=tf.float32)
diff_x = tf.concat(axis=2, values=[diff_x_valid, last_col])

diff_y_valid = tf.slice(diff_y, begin=[0, 0, 0, 0],
size=[-1, x.shape[1].value - 1, x.shape[2].value, x.shape[3].value])
last_row = tf.zeros([tf.shape(x)[0], 1, x.shape[2].value, x.shape[3].value], dtype=tf.float32)
diff_y = tf.concat(axis=1, values=[diff_y_valid, last_row])

return diff_x, diff_y

def divergence(self, x, y, name):
assert len(x.shape) == 4

Expand Down Expand Up @@ -310,8 +337,8 @@ def get_loss(self, x1, x2,
max_iterations=max_iterations)

# computing loss
u1x, u1y = self.forward_gradient(u1, 'u1')
u2x, u2y = self.forward_gradient(u2, 'u2')
u1x, u1y = self.forward_gradient_forloss(u1, 'u1')
u2x, u2y = self.forward_gradient_forloss(u2, 'u2')


u1_flat = tf.reshape(u1, (tf.shape(x2)[0], 1, x2.shape[1].value * x2.shape[2].value))
Expand Down