def discriminator(x):
with tf.variable_scope("discriminator", reuse=tf.AUTO_REUSE):
with tf.variable_scope("conv1"):
conv1 = default_conv2d(x, 128)
conv1 = nn.leaky_relu(conv1, alpha=0.2)
with tf.variable_scope("conv2"):
conv2 = default_conv2d(conv1, 256)
conv2 = layers.batch_normalization(conv2)
conv2 = nn.leaky_relu(conv2, alpha=0.2)
with tf.variable_scope("conv3"):
conv3 = default_conv2d(conv2, 512)
conv3 = layers.batch_normalization(conv3)
conv3 = nn.leaky_relu(conv3, alpha=0.2)
with tf.variable_scope("conv4"):
conv4 = default_conv2d(conv3, 1024)
conv4 = layers.batch_normalization(conv3)
conv4 = nn.leaky_relu(conv3, alpha=0.2)
with tf.variable_scope("linear"):
linear = clayers.flatten(conv4)
linear = clayers.fully_connected(linear, 1)
with tf.variable_scope("out"):
out = nn.sigmoid(linear)
return out
def generator_net(inputs, scope, reuse=None, rgb=False): output_channels = 3 if rgb else 1 with tf.variable_scope(scope, reuse=reuse): # branch 1 ( color reconstruction) cv1 = conv2d(inputs, filters=16, kernel_size=3, strides=1, padding='same', activation=None, name='conv9_i') cv1_r = leaky_relu(cv1) res1_c = conv2d(cv1_r, filters=16, kernel_size=5, strides=1, padding='same', activation=None, name='conv3a_1') res1_b = batch_normalization(res1_c) res1_r = leaky_relu(res1_b) res1_d = conv2d(res1_r, filters=16, kernel_size=3, strides=1, padding='same', activation=None, name='conv3b_1') res1 = batch_normalization(res1_d) sum1 = cv1 + res1 res2_c = conv2d(sum1, filters=16, kernel_size=3, strides=1, padding='same', activation=None, name='conv3a_2') res2_b = batch_normalization(res2_c) res2_r = leaky_relu(res2_b) res2_d = conv2d(res2_r, filters=16, kernel_size=3, strides=1, padding='same', activation=None, name='conv3b_2') res2 = batch_normalization(res2_d) br1 = sum1 + res2 # branch 2 (features extraction) br2 = conv2d(inputs, filters=16, kernel_size=5, strides=1, padding='same', activation=tf.nn.leaky_relu, name='conv_bf1') br2 = max_pooling2d(br2, pool_size=2, strides=2, name='maxpool1') br2 = conv2d(br2, filters=16, kernel_size=3, strides=1, padding='same', activation=tf.nn.leaky_relu, name='conv_bf2') br2 = max_pooling2d(br2, pool_size=2, strides=2, name='maxpool2a') br2 = conv2d(br2, filters=16, kernel_size=3, strides=1, padding='same', activation=tf.nn.leaky_relu, name='conv_bf3') br2 = max_pooling2d(br2, pool_size=2, strides=2, name='maxpool2') print(br2.shape) br2 = conv2d_transpose(br2, filters=16, kernel_size=3, padding='same', strides=2, activation=tf.nn.leaky_relu, name="deconv_1") print(br2.shape) br2 = conv2d_transpose(br2, filters=16, kernel_size=3, padding='same', strides=2, activation=tf.nn.leaky_relu, name="deconv_2") print(br2.shape) br2 = conv2d_transpose(br2, filters=16, kernel_size=3, padding='same', strides=2, activation=tf.nn.leaky_relu, name="deconv_3") print(br2.shape) # concatenate branches and reconstruct image sum3 = tf.concat((br1, br2), axis=3); model = conv2d(sum3, filters=output_channels, kernel_size=3, strides=1, padding='same', activation=None, name='conv9_f') return model
def conv_layer(x,
filtershape,
stride,
name,
LReLU=True,
BN=True,
IS_Training=True):
with tf.variable_scope(name):
filters = tf.get_variable(name='weight',
shape=filtershape,
dtype=tf.float32,
initializer=tf.random_normal_initializer(
mean=0, stddev=0.2),
trainable=True)
conv = tf.nn.conv2d(x, filters, [1, stride, stride, 1], padding='SAME')
conv_biases = tf.Variable(tf.constant(0.1,
shape=[filtershape[3]],
dtype=tf.float32),
trainable=True,
name='bias')
bias = tf.nn.bias_add(conv, conv_biases)
output = bias
if (BN == True):
output = bn(output, IS_Training, name + '_bn')
if (LReLU == True):
output = nn.leaky_relu(bias)
#output = tf.nn.dropout(prelu, keep_prob=keep_prob)
#img_filt = tf.reshape(filters[:,:,:,1], [-1,filtershape[0],filtershape[1],1])
#tf.summary.image('conv_filter',img_filt)
return output
def ACT(inputs, act_fn):
if act_fn == 'relu':
act = relu(inputs)
elif act_fn == 'lrelu':
act = leaky_relu(inputs)
elif act_fn == 'sigmoid':
act = sigmoid(inputs)
return act
def leaky_relu(inputs, alpha=0.2):
"""
Leaky relu activation
Parameters
----------
inputs: Input tensor
alpha: Slope of negative neurons
"""
return nn.leaky_relu(inputs, alpha=alpha)
def convBatchNorm(num,
x,
num_filters,
filter_size=5,
strides=(2, 2),
training=True):
with tf.variable_scope('conv-batch-norm-{}'.format(num)):
x = conv2d(x, num_filters, filter_size, strides=strides)
x = batch_norm(x, training=training)
x = leaky_relu(x)
return x
def __init__(self, images, params, reuse=False):
with tf.variable_scope("RGBNetwork", reuse=reuse) as scope:
x = tf.reshape(images, [-1, 200, 200, 3])
x = conv2d(x, 64, (10, 10), strides=(4, 4), padding='same')
x = leaky_relu(x)
x = max_pooling2d(x, (5, 5), strides=(2, 2), padding='same')
x = conv2d(x, 128, (2, 2), strides=(2, 2), padding='same')
x = leaky_relu(x)
x = max_pooling2d(x, (2, 2), strides=(2, 2), padding='same')
x = conv2d(x, 256, (3, 3), strides=(1, 1), padding='same')
x = leaky_relu(x)
x = conv2d(x, 512, (3, 3), strides=(1, 1), padding='same')
x = leaky_relu(x)
x = flatten(x)
x = dense(x, 256, activation=leaky_relu)
x = dense(x, 200 * 200, activation=None)
x = tf.reshape(x, [-1, 200, 200])
self.output = tf.cast(x, tf.float32)
self.parameters = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,
scope="SimpleNetwork")
def ACT(inputs, act_fn):
if act_fn == 'relu':
act = relu(inputs)
elif act_fn == 'lrelu':
act = leaky_relu(inputs)
elif act_fn == 'sigmoid':
act = sigmoid(inputs)
elif act_fn == 'tanh':
act = tanh(inputs)
else:
act = inputs
return act
def convolutional_autoencoder_net(inputs, scope, reuse=None, rgb=False): output_channels = 3 if rgb else 1 with tf.variable_scope(scope, reuse=reuse): cv1 = conv2d(inputs, filters=32, kernel_size=3, strides=1, padding='same', activation=None, name='conv9_i') cv1_r = leaky_relu(cv1) res1_c = conv2d(cv1_r, filters=32, kernel_size=5, strides=1, padding='same', activation=None, name='conv3a_1') res1_b = batch_normalization(res1_c) res1_r = leaky_relu(res1_b) res1_d = conv2d(res1_r, filters=32, kernel_size=3, strides=1, padding='same', activation=None, name='conv3b_1') res1 = batch_normalization(res1_d) sum1 = cv1 + res1 res2_c = conv2d(sum1, filters=32, kernel_size=3, strides=1, padding='same', activation=None, name='conv3a_2') res2_b = batch_normalization(res2_c) res2_r = leaky_relu(res2_b) res2_d = conv2d(res2_r, filters=32, kernel_size=5, strides=1, padding='same', activation=None, name='conv3b_2') res2 = batch_normalization(res2_d) sum2 = sum1 + res2 res3_c = conv2d(sum2, filters=32, kernel_size=3, strides=1, padding='same', activation=None, name='conv3a_3') res3_b = batch_normalization(res3_c) res3_r = leaky_relu(res3_b) res3_d = conv2d(res3_r, filters=32, kernel_size=3, strides=1, padding='same', activation=None, name='conv3b_3') res3 = batch_normalization(res3_d) sum3 = sum2 + res3 model = conv2d(sum3, filters=output_channels, kernel_size=3, strides=1, padding='same', activation=None, name='conv9_f') return model
def predict(self, inputTensor, _):
result = nn.leaky_relu(inputTensor, self.alpha)
return (result)
def relu(x):
return nn.leaky_relu(x)
def _silu(x):
"""(nn.silu())"""
return x * tf.sigmoid(x)
x = tf.constant([[-1., 0, 1], [0, 0, 1]])
print(_relu(x))
print(nn.relu(x))
# tf.Tensor(
# [[0. 0. 1.]
# [0. 0. 1.]], shape=(2, 3), dtype=float32)
# tf.Tensor(
# [[0. 0. 1.]
# [0. 0. 1.]], shape=(2, 3), dtype=float32)
print(_leaky_relu(x))
print(nn.leaky_relu(x))
# tf.Tensor(
# [[-0.2 0. 1. ]
# [ 0. 0. 1. ]], shape=(2, 3), dtype=float32)
# tf.Tensor(
# [[-0.2 0. 1. ]
# [ 0. 0. 1. ]], shape=(2, 3), dtype=float32)
print(_sigmoid(x))
print(nn.sigmoid(x))
# tf.Tensor(
# [[0.26894143 0.5 0.73105854]
# [0.5 0.5 0.73105854]], shape=(2, 3), dtype=float32)
# tf.Tensor(
# [[0.26894143 0.5 0.73105854]
# [0.5 0.5 0.7310586 ]], shape=(2, 3), dtype=float32)
print(_tanh(x))
def _GuidedReluGrad(op, grad):
return tf.where(0. < grad, nn.leaky_relu(grad, op.outputs[0]),
tf.zeros_like(grad))
