def discriminator(x, is_training=True, reuse=None):
with tf.variable_scope("discriminator", reuse=reuse) as scope:
# Variables for classifier
D_W1 = utils.weight_variable([4, 4, 3, K], name="D_W1")
D_B1 = utils.bias_variable([K], name="D_B1")
D_W2 = utils.weight_variable([4, 4, K, L], name="D_W2")
D_B2 = utils.bias_variable([L], name="D_B2")
D_W3 = utils.weight_variable([8 * 8 * L, M], name="D_W3")
D_B3 = utils.bias_variable([M], name="D_B3")
D_W4 = utils.weight_variable([M, 1], name="D_W4")
D_B4 = utils.bias_variable([1], name="D_B4")
stride = 2 # output is 16x16
H1 = lrelu(
tf.nn.conv2d(
x, D_W1, strides=[1, stride, stride, 1], padding='SAME') +
D_B1)
print(H1.shape)
stride = 2 # output is 8x8
H2 = lrelu(
utils.bn((tf.nn.conv2d(
H1, D_W2, strides=[1, stride, stride, 1], padding='SAME') +
D_B2),
is_training=is_training,
scope="D_bn_h2"))
print(H2.shape)
# reshape the output from the third convolution for the fully connected layer
HH2 = tf.reshape(H2, shape=[-1, 8 * 8 * L])
H3 = lrelu(tf.matmul(HH2, D_W3) + D_B3)
Ylogits = tf.matmul(H3, D_W4) + D_B4
Ysigmoid = tf.nn.sigmoid(Ylogits)
Ysoftmax = tf.nn.softmax(Ylogits)
return Ysoftmax, Ysigmoid, Ylogits
X_noisy = tf.placeholder(tf.float32, [None, 28, 28, 1])
X_adv = tf.placeholder(tf.float32, [None, 28, 28, 1])
GX_ = tf.placeholder(tf.float32, [None, 28, 28, 1])
# output Y_ & input GY: labels for classification and generation
Y_ = tf.placeholder(tf.float32, [None, 10])
GY = tf.placeholder(tf.float32, [None, 10])
input_test_sum = tf.summary.image("input", X, 10)
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, 10)
input_adv_sum = tf.summary.image("input-adv", X_adv, 10)
with tf.name_scope("classifier-generator"):
# Weights for classifier and generator
C_W1 = utils.weight_variable([784, L], stddev=0.1, name="C_W1")
C_W2 = utils.weight_variable([L, M], stddev=0.1, name="C_W2")
C_W3 = utils.weight_variable([M, 10], stddev=0.1, name="C_W3")
def classifier(x, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([L], name="C_B1")
C_B2 = utils.bias_variable([M], name="C_B2")
C_B3 = utils.bias_variable([10], name="C_B3")
XX = tf.reshape(x, [-1, 784])
H1 = tf.nn.sigmoid(tf.matmul(XX, C_W1) + C_B1)
H2 = tf.nn.sigmoid(tf.matmul(H1, C_W2) + C_B2)
Ylogits = tf.matmul(H2, C_W3) + C_B3
X_adv = tf.placeholder(tf.float32, [None, 32, 32, 3])
GX_ = tf.placeholder(tf.float32, [None, 32, 32, 3])
# output Y_ & input GY: labels for classification and generation
Y_ = tf.placeholder(tf.float32, [None, num_classes])
GY = tf.placeholder(tf.float32, [None, num_classes])
input_test_sum = tf.summary.image("input", X, num_classes)
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, num_classes)
input_adv_sum = tf.summary.image("input-adv", X_adv, num_classes)
with tf.name_scope("classifier-generator"):
# Weights for classifier and generator
C_W1 = utils.weight_variable([32 * 32 * 3, num_classes],
stddev=0.1,
name="C_W1")
def classifier(x, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([num_classes], name="C_B1")
XX = tf.reshape(x, [-1, 32 * 32 * 3])
Ylogits = tf.matmul(XX, C_W1) + C_B1
Ysigmoid = tf.nn.sigmoid(Ylogits)
Ysoftmax = tf.nn.softmax(Ylogits)
return Ysoftmax, Ysigmoid, Ylogits
X_noisy = tf.placeholder(tf.float32, [None, 32, 32, 3])
X_adv = tf.placeholder(tf.float32, [None, 32, 32, 3])
GX_ = tf.placeholder(tf.float32, [None, 32, 32, 3])
# output Y_ & input GY: labels for classification and generation
Y_ = tf.placeholder(tf.float32, [None, num_classes])
GY = tf.placeholder(tf.float32, [None, num_classes])
input_test_sum = tf.summary.image("input", X, num_classes)
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, num_classes)
input_adv_sum = tf.summary.image("input-adv", X_adv, num_classes)
with tf.name_scope("classifier-generator"):
# Weights for classifier and generator
C_W1 = utils.weight_variable([32 * 32 * 3, L], stddev=0.1, name="C_W1")
C_W2 = utils.weight_variable([L, M], stddev=0.1, name="C_W2")
C_W3 = utils.weight_variable([M, N], stddev=0.1, name="C_W3")
C_W4 = utils.weight_variable([N, O], stddev=0.1, name="C_W4")
C_W5 = utils.weight_variable([O, num_classes], stddev=0.1, name="C_W5")
def classifier(x, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([L], name="C_B1")
C_B2 = utils.bias_variable([M], name="C_B2")
C_B3 = utils.bias_variable([N], name="C_B3")
C_B4 = utils.bias_variable([O], name="C_B4")
C_B5 = utils.bias_variable([num_classes], name="C_B5")
# output Y_ & input GY: labels for classification and generation
Y_ = tf.placeholder(tf.float32, [None, 10])
GY = tf.placeholder(tf.float32, [None, 10])
# variable learning rate
lr = tf.placeholder(tf.float32)
# variable batch size
BS = tf.placeholder(tf.int32)
input_test_sum = tf.summary.image("input", X, 10)
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, 10)
input_adv_sum = tf.summary.image("input-adv", X_adv, 10)
with tf.name_scope("classifier-generator"):
C_W1 = utils.weight_variable([5, 5, 1, K], stddev=0.1, name="C_W1")
C_W2 = utils.weight_variable([5, 5, K, L], stddev=0.1, name="C_W2")
C_W3 = utils.weight_variable([4, 4, L, M], stddev=0.1, name="C_W3")
C_W4 = utils.weight_variable([7 * 7 * M, N], stddev=0.1, name="C_W4")
C_W5 = utils.weight_variable([N, 10], stddev=0.1, name="C_W5")
def classifier(x, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([K], name="C_B1")
C_B2 = utils.bias_variable([L], name="C_B2")
C_B3 = utils.bias_variable([M], name="C_B3")
C_B4 = utils.bias_variable([N], name="C_B4")
C_B5 = utils.bias_variable([10], name="C_B5")
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, num_classes)
input_adv_sum = tf.summary.image("input-adv", X_adv, num_classes)
def sample_Z(m, n):
return np.random.uniform(-1., 1., size=[m, n])
# From tensorflow-generative-model-collections
def lrelu(x, leak=0.2):
return tf.maximum(x, leak * x)
with tf.name_scope("classifier-generator"):
# Weights for classifier and generator
C_W1 = utils.weight_variable([4, 4, 3, K], name="C_W1")
C_W2 = utils.weight_variable([4, 4, K, L], name="C_W2")
C_W3 = utils.weight_variable([8 * 8 * L, M], name="C_W3")
C_W4 = utils.weight_variable([M, num_classes], name="C_W4")
def classifier(x, is_training=True, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([K], name="C_B1")
C_B2 = utils.bias_variable([L], name="C_B2")
C_B3 = utils.bias_variable([M], name="C_B3")
C_B4 = utils.bias_variable([num_classes], name="C_B4")
stride = 2 # output is 16x16
X_noisy = tf.placeholder(tf.float32, [None, 28, 28, 1])
X_adv = tf.placeholder(tf.float32, [None, 28, 28, 1])
GX_ = tf.placeholder(tf.float32, [None, 28, 28, 1])
# output Y_ & input GY: labels for classification and generation
Y_ = tf.placeholder(tf.float32, [None, 10])
GY = tf.placeholder(tf.float32, [None, 10])
input_test_sum = tf.summary.image("input", X, 10)
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, 10)
input_adv_sum = tf.summary.image("input-adv", X_adv, 10)
with tf.name_scope("classifier-generator"):
# Weights for classifier and generator
C_W1 = utils.weight_variable([784, 10], stddev=0.1, name="C_W1")
def classifier(x, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([10], name="C_B1")
XX = tf.reshape(x, [-1, 784])
Ylogits = tf.matmul(XX, C_W1) + C_B1
Ysigmoid = tf.nn.sigmoid(Ylogits)
Ysoftmax = tf.nn.softmax(Ylogits)
return Ysoftmax, Ysigmoid, Ylogits
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, 10)
input_adv_sum = tf.summary.image("input-adv", X_adv, 10)
def sample_Z(m, n):
return np.random.uniform(-1., 1., size=[m, n])
# From tensorflow-generative-model-collections
def lrelu(x, leak=0.2):
return tf.maximum(x, leak * x)
with tf.name_scope("classifier-generator"):
# Weights for classifier and generator
C_W1 = utils.weight_variable([4, 4, 1, K], name="C_W1")
C_W2 = utils.weight_variable([4, 4, K, L], name="C_W2")
C_W3 = utils.weight_variable([7 * 7 * L, M], name="C_W3")
C_W4 = utils.weight_variable([M, Z_dim], name="C_W4")
def classifier(x, is_training=True, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([K], name="C_B1")
C_B2 = utils.bias_variable([L], name="C_B2")
C_B3 = utils.bias_variable([M], name="C_B3")
C_B4 = utils.bias_variable([Z_dim], name="C_B4")
stride = 2 # output is 14x14
X_noisy = tf.placeholder(tf.float32, [None, 28, 28, 1])
X_adv = tf.placeholder(tf.float32, [None, 28, 28, 1])
GX_ = tf.placeholder(tf.float32, [None, 28, 28, 1])
# output Y_ & input GY: labels for classification and generation
Y_ = tf.placeholder(tf.float32, [None, 10])
GY = tf.placeholder(tf.float32, [None, 10])
input_test_sum = tf.summary.image("input", X, 10)
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, 10)
input_adv_sum = tf.summary.image("input-adv", X_adv, 10)
with tf.name_scope("classifier-generator"):
# Weights for classifier and generator
C_W1 = utils.weight_variable([784, L], stddev=0.1, name="C_W1")
C_W2 = utils.weight_variable([L, M], stddev=0.1, name="C_W2")
C_W3 = utils.weight_variable([M, N], stddev=0.1, name="C_W3")
C_W4 = utils.weight_variable([N, O], stddev=0.1, name="C_W4")
C_W5 = utils.weight_variable([O, 10], stddev=0.1, name="C_W5")
def classifier(x, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([L], name="C_B1")
C_B2 = utils.bias_variable([M], name="C_B2")
C_B3 = utils.bias_variable([N], name="C_B3")
C_B4 = utils.bias_variable([O], name="C_B4")
C_B5 = utils.bias_variable([10], name="C_B5")
X_noisy = tf.placeholder(tf.float32, [None, 32, 32, 3])
X_adv = tf.placeholder(tf.float32, [None, 32, 32, 3])
GX_ = tf.placeholder(tf.float32, [None, 32, 32, 3])
# output Y_ & input GY: labels for classification and generation
Y_ = tf.placeholder(tf.float32, [None, num_classes])
GY = tf.placeholder(tf.float32, [None, num_classes])
input_test_sum = tf.summary.image("input", X, num_classes)
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, num_classes)
input_adv_sum = tf.summary.image("input-adv", X_adv, num_classes)
with tf.name_scope("classifier-generator"):
# Weights for classifier and generator
C_W1 = utils.weight_variable([32 * 32 * 3, L], stddev=0.1, name="C_W1")
C_W2 = utils.weight_variable([L, M], stddev=0.1, name="C_W2")
C_W3 = utils.weight_variable([M, num_classes], stddev=0.1, name="C_W3")
def classifier(x, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([L], name="C_B1")
C_B2 = utils.bias_variable([M], name="C_B2")
C_B3 = utils.bias_variable([num_classes], name="C_B3")
XX = tf.reshape(x, [-1, 32 * 32 * 3])
H1 = tf.nn.sigmoid(tf.matmul(XX, C_W1) + C_B1)
H2 = tf.nn.sigmoid(tf.matmul(H1, C_W2) + C_B2)
Ylogits = tf.matmul(H2, C_W3) + C_B3
# output Y_ & input GY: labels for classification and generation
Y_ = tf.placeholder(tf.float32, [None, num_classes])
GY = tf.placeholder(tf.float32, [None, num_classes])
# variable learning rate
lr = tf.placeholder(tf.float32)
# variable batch size
BS = tf.placeholder(tf.int32)
input_test_sum = tf.summary.image("input", X, num_classes)
input_noisy_sum = tf.summary.image("input-noisy", X_noisy, num_classes)
input_adv_sum = tf.summary.image("input-adv", X_adv, num_classes)
with tf.name_scope("classifier-generator"):
C_W1 = utils.weight_variable([5, 5, 3, K], stddev=0.1, name="C_W1")
C_W2 = utils.weight_variable([5, 5, K, L], stddev=0.1, name="C_W2")
C_W3 = utils.weight_variable([4, 4, L, M], stddev=0.1, name="C_W3")
C_W4 = utils.weight_variable([8 * 8 * M, N], stddev=0.1, name="C_W4")
C_W5 = utils.weight_variable([N, num_classes], stddev=0.1, name="C_W5")
def classifier(x, reuse=None):
with tf.variable_scope("classifier", reuse=reuse) as scope_c:
# Variables for classifier
C_B1 = utils.bias_variable([K], name="C_B1")
C_B2 = utils.bias_variable([L], name="C_B2")
C_B3 = utils.bias_variable([M], name="C_B3")
C_B4 = utils.bias_variable([N], name="C_B4")
C_B5 = utils.bias_variable([num_classes], name="C_B5")