def build_arch(self):
with tf.variable_scope('Conv1_layer'):
# Conv1, [batch, 20, 20, 256]
conv1 = tf.contrib.layers.conv2d(self.X,
num_outputs=256,
kernel_size=9,
stride=1,
padding='VALID')
assert conv1.get_shape() == [cfg.batch_size, 20, 20, 256]
# Primary Capsules, [batch_size, 1152, 8, 1]
with tf.variable_scope('Primary_Caps_layer'):
primaryCaps = CapsConv(num_units=8, with_routing=False)
caps1 = primaryCaps(conv1, num_outputs=32, kernel_size=9, stride=2)
assert caps1.get_shape() == [cfg.batch_size, 1152, 8, 1]
# DigitCaps layer, [batchj_size, 10, 16, 1]
with tf.variable_scope('DigitCaps_layer'):
digitCaps = CapsConv(num_units=16, with_routing=True)
self.caps2 = digitCaps(caps1, num_outputs=10)
# 1. Masking
with tf.variable_scope('Masking'):
# a). calc ||v_c||, then do softmax(||v_c||)
# [batch_size, 10, 16, 1] => [batch_size, 10, 1, 1]
self.v_length = tf.sqrt(
tf.reduce_sum(tf.square(self.caps2), axis=2, keep_dims=True))
self.softmax_v = tf.nn.softmax(self.v_length, dim=1)
assert self.softmax_v.get_shape() == [cfg.batch_size, 10, 1, 1]
# b). pick out the index of the max softmax val of the 10 caps
# [batch_size, 10, 1, 1] => [batch_size] (index)
argmax_idx = tf.argmax(self.softmax_v,
axis=1,
output_type=tf.int32)
assert argmax_idx.get_shape() == [cfg.batch_size, 10, 1, 1]
# c). indexing
masked_v = []
argmax_idx = tf.reshape(argmax_idx, shape=(cfg.batch_size, ))
for batch_size in range(cfg.batch_size):
v = self.caps2[batch_size][argmax_idx[batch_size], :]
masked_v.append(tf.reshape(v, shape=(1, 1, 16, 1)))
self.masked_v = tf.concat(masked_v, axis=0)
assert self.masked_v.get_shape() == [cfg.batch_size, 1, 16, 1]
# 2. reconstruct the MNIST images with 3 FC layers
# [batch_size, 1, 16, 1] => [batch_size, 16] => [batch_size, 512]
with tf.variable_scope('Decoder'):
vector_j = tf.reshape(self.masked_v, shape=(cfg.batch_size, -1))
fc1 = tf.contrib.layers.fully_connected(vector_j, num_outputs=512)
assert fc1.get_shape() == [cfg.batch_size, 512]
fc2 = tf.contrib.layers.fully_connected(fc1, num_outputs=1024)
assert fc2.get_shape() == [cfg.batch_size, 1024]
self.decoded = tf.contrib.layers.fully_connected(
fc2, num_outputs=784, activation_fn=tf.sigmoid)
def build_arch(self):
with tf.variable_scope('Conv1_layer'):
# Conv1, [batch_size, 20, 20, 256]
conv1 = tf.contrib.layers.conv2d(self.X, num_outputs=256,
kernel_size=9, stride=1,
padding='VALID')
assert conv1.get_shape() == [cfg.batch_size, 20, 20, 256]
# TODO: Rewrite the 'CapsConv' class as a function, the capsLay
# function should be encapsulated into two functions, one like conv2d
# and another is fully_connected in Tensorflow.
# Primary Capsules, [batch_size, 1152, 8, 1]
with tf.variable_scope('PrimaryCaps_layer'):
primaryCaps = CapsConv(num_units=8, with_routing=False)
caps1 = primaryCaps(conv1, num_outputs=32, kernel_size=9, stride=2)
assert caps1.get_shape() == [cfg.batch_size, 1152, 8, 1]
# DigitCaps layer, [batch_size, 10, 16, 1]
with tf.variable_scope('DigitCaps_layer'):
digitCaps = CapsConv(num_units=16, with_routing=True)
self.caps2 = digitCaps(caps1, num_outputs=10)
# Decoder structure in Fig. 2
# 1. Do masking, how:
with tf.variable_scope('Masking'):
# a). calc ||v_c||, then do softmax(||v_c||)
# [batch_size, 10, 16, 1] => [batch_size, 10, 1, 1]
self.v_length = tf.sqrt(tf.reduce_sum(tf.square(self.caps2),
axis=2, keep_dims=True))
self.softmax_v = tf.nn.softmax(self.v_length, dim=1)
assert self.softmax_v.get_shape() == [cfg.batch_size, 10, 1, 1]
# b). pick out the index of max softmax val of the 10 caps
# [batch_size, 10, 1, 1] => [batch_size] (index)
argmax_idx = tf.argmax(self.softmax_v, axis=1, output_type=tf.int32)
assert argmax_idx.get_shape() == [cfg.batch_size, 1, 1]
# c). indexing
# It's not easy to understand the indexing process with argmax_idx
# as we are 3-dim animal
masked_v = []
argmax_idx = tf.reshape(argmax_idx, shape=(cfg.batch_size, ))
for batch_size in range(cfg.batch_size):
v = self.caps2[batch_size][argmax_idx[batch_size], :]
masked_v.append(tf.reshape(v, shape=(1, 1, 16, 1)))
self.masked_v = tf.concat(masked_v, axis=0)
assert self.masked_v.get_shape() == [cfg.batch_size, 1, 16, 1]
# 2. Reconstructe the MNIST images with 3 FC layers
# [batch_size, 1, 16, 1] => [batch_size, 16] => [batch_size, 512]
with tf.variable_scope('Decoder'):
vector_j = tf.reshape(self.masked_v, shape=(cfg.batch_size, -1))
fc1 = tf.contrib.layers.fully_connected(vector_j, num_outputs=512)
assert fc1.get_shape() == [cfg.batch_size, 512]
fc2 = tf.contrib.layers.fully_connected(fc1, num_outputs=1024)
assert fc2.get_shape() == [cfg.batch_size, 1024]
self.decoded = tf.contrib.layers.fully_connected(fc2, num_outputs=784, activation_fn=tf.sigmoid)
def build_arch(self):
with tf.variable_scope('Conv1_layer'):
# Conv1, [batch_size, 20, 20, 256]
conv1 = tf.contrib.layers.conv2d(self.X,
num_outputs=256,
kernel_size=9,
stride=1,
padding='VALID')
assert conv1.get_shape() == [cfg.batch_size, 20, 20, 256]
# TODO: Rewrite the 'CapsConv' class as a function
# Primary Capsules, [batch_size, 1152, 8, 1]
with tf.variable_scope('PrimaryCaps_layer'):
primaryCaps = CapsConv(num_units=8, with_routing=False)
caps1 = primaryCaps(conv1, num_outputs=32, kernel_size=9, stride=2)
assert caps1.get_shape() == [cfg.batch_size, 1152, 8, 1]
# DigitCaps layer, [batch_size, 10, 16, 1]
with tf.variable_scope('DigitCaps_layer'):
digitCaps = CapsConv(num_units=16, with_routing=True)
self.caps2 = digitCaps(caps1, num_outputs=10)
with tf.variable_scope('Masking'):
# [batch_size, 10, 16, 1] => [batch_size, 10, 1, 1]
self.v_length = tf.sqrt(
tf.reduce_sum(tf.square(self.caps2), axis=2, keep_dims=True))
self.softmax_v = tf.nn.softmax(self.v_length, dim=1)
assert self.softmax_v.get_shape() == [cfg.batch_size, 10, 1, 1]
# [batch_size, 10, 1, 1] => [batch_size] (index)
argmax_idx = tf.argmax(self.softmax_v,
axis=1,
output_type=tf.int32)
assert argmax_idx.get_shape() == [cfg.batch_size, 1, 1]
# as we are 3-dim animal
masked_v = []
argmax_idx = tf.reshape(argmax_idx, shape=(cfg.batch_size, ))
for batch_size in range(cfg.batch_size):
v = self.caps2[batch_size][argmax_idx[batch_size], :]
masked_v.append(tf.reshape(v, shape=(1, 1, 16, 1)))
self.masked_v = tf.concat(masked_v, axis=0)
assert self.masked_v.get_shape() == [cfg.batch_size, 1, 16, 1]
# [batch_size, 1, 16, 1] => [batch_size, 16] => [batch_size, 512]
with tf.variable_scope('Decoder'):
vector_j = tf.reshape(self.masked_v, shape=(cfg.batch_size, -1))
fc1 = tf.contrib.layers.fully_connected(vector_j, num_outputs=512)
assert fc1.get_shape() == [cfg.batch_size, 512]
fc2 = tf.contrib.layers.fully_connected(fc1, num_outputs=1024)
assert fc2.get_shape() == [cfg.batch_size, 1024]
self.decoded = tf.contrib.layers.fully_connected(
fc2, num_outputs=784, activation_fn=tf.sigmoid)
def build_arch(self):
# Conv1
conv1 = tf.contrib.layers.conv2d(self.input,
num_outputs=256,
kernel_size=9,
strides=1)
# Primary Capsules
primaryCaps = CapsConv(num_units=8)
caps1 = primaryCaps(conv1, num_outputs=32, kernel_size=9, strides=2)
# DigitCaps layer
digitCaps = CapsConv(num_units=16)
caps2 = digitCaps(caps1, num_outputs=10, kernel_size=9, strides=2)
# Decoder structure in Fig. 2
# TODO: before reconstruction the input caps2 should do masking to pick
# out the activity vector of the correct digit capsule.
fc1 = tf.contrib.layers.fully_connected(caps2, num_outputs=512)
fc2 = tf.contrib.layers.fully_connected(fc1, num_outputs=1024)
self.decoded = tf.contrib.layers.fully_connected(fc2, num_outputs=784)
def build_arch(self):
# Conv1, [batch_size, 20, 20, 256]
conv1 = tf.contrib.layers.conv2d(self.X,
num_outputs=256,
kernel_size=9,
stride=1)
# Primary Capsules, [batch_size, 1152, 8, 1]
primaryCaps = CapsConv(num_units=8, with_routing=False)
caps1 = primaryCaps(conv1, num_outputs=32, kernel_size=9, strides=2)
# DigitCaps layer, [batch_size, 10, 16, 1]
digitCaps = CapsConv(num_units=16, with_routing=True)
self.caps2 = digitCaps(caps1, num_outputs=10)
# Decoder structure in Fig. 2
# TODO: before reconstruction the input caps2 should do masking to pick
# out the activity vector of the correct digit capsule.
fc1 = tf.contrib.layers.fully_connected(self.caps2, num_outputs=512)
fc2 = tf.contrib.layers.fully_connected(fc1, num_outputs=1024)
self.decoded = tf.contrib.layers.fully_connected(fc2, num_outputs=784)