def model(this,
inputs,
is_training,
num_outputs=[None, 128, 64, 32],
**kwargs):
with tf.variable_scope('feature_fusion', values=kwargs["values"]):
with this.get_slime_arg_scope(is_training,
weight_decay=kwargs["weight_decay"]):
g = []
h = []
for i in range(len(num_outputs)):
print('Shape of f_{} {}'.format(i, inputs[i].shape))
g.append(None)
h.append(None)
for i in range(len(num_outputs)):
if i == 0:
h[i] = inputs[i]
else:
c1_1 = slim.conv2d(
tf.concat([g[i - 1], inputs[i]], axis=-1),
num_outputs[i], 1)
h[i] = slim.conv2d(c1_1, num_outputs[i], 3)
if i <= len(num_outputs) - 2:
g[i] = dogoo_ops.unpool(h[i])
else:
g[i] = slim.conv2d(h[i], num_outputs[i], 3)
print('Shape of h_{} {}, g_{} {}'.format(
i, h[i].shape, i, g[i].shape))
return g
def model(this,
inputs,
is_training,
num_outputs=[128, 128, 64, 32],
dabcids=[],
**kwargs):
num_basis = [8, 4, 4, 4]
with tf.variable_scope('feature_fusion', values=kwargs["values"]):
with this.get_slime_arg_scope(is_training,
weight_decay=kwargs["weight_decay"]):
g = []
h = []
l = []
w = []
for i in range(len(num_outputs)):
print('Shape of f_{} {}'.format(i, inputs[i].shape))
g.append(None)
h.append(None)
l.append(None)
w.append(None)
for i in range(len(num_outputs)):
if i == 0:
if (4 - i not in dabcids):
w[i], l[i] = this.cat_coabc(
inputs[i], inputs[i], num_basis[i],
num_outputs[i])
else:
print("disabling ", str(4 - i))
w[i], l[i] = this.cat_dcoabc(
inputs[i], inputs[i], num_basis[i],
num_outputs[i])
h[i] = l[i]
else:
if (4 - i not in dabcids):
w[i], l[i] = this.cat_coabc(
inputs[i], g[i - 1], num_basis[i],
num_outputs[i])
else:
print("disabling ", str(4 - i))
w[i], l[i] = this.cat_dcoabc(
inputs[i], g[i - 1], num_basis[i],
num_outputs[i])
c1_1 = slim.conv2d(
tf.concat([g[i - 1], l[i]], axis=-1),
num_outputs[i], 1)
h[i] = slim.conv2d(c1_1, num_outputs[i], 3)
if i <= len(num_outputs) - 2:
g[i] = dogoo_ops.unpool(h[i])
else:
g[i] = slim.conv2d(h[i], num_outputs[i], 3)
print('Shape of h_{} {}, g_{} {}'.format(
i, h[i].shape, i, g[i].shape))
return g