def create_model(self, x):
print('0', x.shape)
x = Conv(x,
filter=2 * self.filters,
kernel=conv0_weight,
stride=conv0_stride,
layer_name='conv0')
print('1', x.shape)
x = Max_Pool(x)
print('2', x.shape)
for i in range(self.num_blocks):
x = self.dense_block(x,
num_layers=num_layers_1,
layer_name='dense_%d' % i)
print('3', x.shape)
x = self.transition(x, scope='trans_%d' % i)
print('4', x.shape)
x = self.dense_block_last(x,
num_layers=num_layers_last,
layer_name='dense_final')
print('5', x.shape)
x = BN(x, training=self.training, scope='linear_batch')
x = lrelu(x)
x = tf.reshape(x, shape=[-1, ts])
print('6', x.shape)
return x
def bottleneck_last(self, x, scope):
with tf.name_scope(scope):
x = BN(x, training=self.training, scope=scope + '_batch1_last')
x = lrelu(x)
x = Conv(x,
filter=4 * self.filters_last,
kernel=[1, 1],
layer_name=scope + '_conv1_last')
# x = Dropout(x, rate=dropout_rate, training=self.training)
x = BN(x, training=self.training, scope=scope + '_batch2')
x = lrelu(x)
x = Conv(x,
filter=self.filters_last,
kernel=conv_weight,
layer_name=scope + '_conv2_last')
# x = Dropout(x, rate=dropout_rate, training=self.training)
return x
def transition(self, x, scope):
with tf.name_scope(scope):
x = BN(x, training=self.training, scope=scope + '_batch1')
x = lrelu(x)
x = Conv(x,
filter=self.filters,
kernel=[1, 1],
layer_name=scope + '_conv1')
# x = Dropout(x, rate=dropout_rate, training=self.training)
x = Avg_Pool(x)
return x
def create_model(self, x):
print('0', x.shape)
x = Conv(x,
filter=2 * self.filters,
kernel=conv0_weight,
stride=conv0_stride,
layer_name='conv0')
print('1', x.shape)
x = Max_Pool(x)
print('2', x.shape)
for i in range(self.num_blocks):
x = self.dense_block(x,
num_layers=num_layers_1,
layer_name='dense_%d' % i)
print('3', x.shape)
x = self.transition(x, scope='trans_%d' % i)
print('4', x.shape)
x = self.dense_block_last(x,
num_layers=num_layers_last,
layer_name='dense_final')
print('5', x.shape)
x = BN(x, training=self.training, scope='linear_batch')
x = lrelu(x)
x = tf.reshape(x, shape=[-1, ts])
print('6', x.shape)
y_final_prime2 = x
y_final_ = tf.slice(x, [0, 0], [-1, 95])
y_final_prime1 = tf.concat([x_init, y_final_], 1)
w1_final = tf.Variable(tf.random_normal([96, 96]))
y_pred = tf.add(tf.matmul(y_final_prime1, w1_final), y_final_prime2)
x = lrelu(y_pred)
return x