def g_tf(args, reuse=False):
eps = 1e-3
"""Tensorflow map from image to pi parameters"""
shapes = [info[port]['shape'] for port in inv_arrow.param_ports()]
inp = args[0]
width, height = getn(options, 'width', 'height')
inp = tf.reshape(inp, (-1, width, height))
inp = tf.expand_dims(inp, axis=3)
from tflearn.layers import conv_2d, fully_connected
tf.summary.image("g_tf_output", inp)
# Do convolutional layers
nlayers = 2
for i in range(nlayers):
inp = conv_2d(inp, nb_filter=4, filter_size=1, activation="elu")
inp = conv_2d(inp,
nb_filter=options['nsteps'],
filter_size=1,
activation="sigmoid")
out = []
for i, shape in enumerate(shapes):
this_inp = inp[:, :, :, i:i + 1]
if shape[1] == width * height:
this_inp = tf.reshape(this_inp,
(options['batch_size'], -1)) + eps
out.append(this_inp)
else:
r_length = int(np.ceil(np.sqrt(shape[1])))
rs = tf.image.resize_images(this_inp, (r_length, r_length))
rs = tf.reshape(rs, (options['batch_size'], -1)) + eps
out.append(rs[:, 0:shape[1]])
return out
def color_transform_layers(inputs):
# convolution 1x1
net = conv_2d(inputs, 20, 1)
net = max_out(net, 10)
net = conv_2d(net, 6, 1)
net = max_out(net, 3)
return net
def conv(classes, input_shape):
model = input_data(input_shape, name="input")
model = conv_2d(model, 32, (3, 3), activation='relu')
model = conv_2d(model, 64, (3, 3), activation='relu')
model = max_pool_2d(model, (2, 2))
model = dropout(model, 0.25)
model = flatten(model)
model = fully_connected(model, 128, activation='relu')
model = dropout(model, 0.5)
model = fully_connected(model, classes, activation='softmax')
model = regression(model,
optimizer='adam',
learning_rate=0.001,
loss='categorical_crossentropy',
name='target')
# Training
model = tflearn.DNN(model, tensorboard_verbose=3)
return model
def conv_2d_layer(input, n_filters, stride):
return conv_2d(input,
n_filters,
3,
strides=stride,
padding='same',
activation='elu',
bias_init='zeros',
scope=None,
name='Conv3D')
def res18_forward(incoming, scope=None, name="resnet_18", reuse=False):
with tf.variable_scope(scope, default_name=name, reuse=reuse):
network = conv_2d(incoming, 32, 5, 2, name="conv1",)
network = residual_block(network, 2, 32, downsample=True, batch_norm=True, name="rb1")
network = residual_block(network, 2, 64, downsample=True, batch_norm=True, name="rb2")
network = residual_block(network, 2, 128, downsample=True, batch_norm=True, name="rb3")
network = residual_block(network, 2, 256, downsample=True, batch_norm=True, name="rb4")
network = relu(batch_normalization(fully_connected(network, 256, name="fc1")))
network = fully_connected(network, 5, name="fc2")
return network
def g_tf(args, reuse=False):
eps = 1e-3
"""Tensorflow map from image to pi parameters"""
shapes = [info[port]['shape'] for port in fwd.in_ports()]
inp = args[0]
width, height = getn(options, 'width', 'height')
inp = tf.reshape(inp, (-1, width, height))
inp = tf.expand_dims(inp, axis=3)
from tflearn.layers import conv_2d, fully_connected
tf.summary.image("g_tf_output", inp)
# Do convolutional layers
nlayers = 2
for i in range(nlayers):
inp = conv_2d(inp, nb_filter=4, filter_size=1, activation="elu")
ratio = width / options['res']
inp = conv_2d(inp,
nb_filter=options['res'],
filter_size=1,
strides=int(ratio),
activation="sigmoid")
return [tf.reshape(inp, (options['batch_size'], -1))]
def code_classifier_forward(config,
incoming=None,
image=None,
scope="code_classifier",
name=None,
reuse=False):
with tf.variable_scope(scope, name, reuse=reuse):
code_output = leaky_relu(fully_connected(incoming, 512))
output = conv_2d(image, 32, 5, 2, name="conv1")
output = residual_block(output,
2,
32,
downsample=True,
batch_norm=True,
name="rb1")
output = residual_block(output,
1,
64,
downsample=True,
batch_norm=True,
name="rb2")
output = leaky_relu(
fully_connected(
tf.reshape(output, [config.batch_size, 4 * 4 * 64]), 1024))
prod = tf.matmul(code_output[:, :, None], output[:, None, :])
prob = tf.nn.softmax(prod)
prob2 = tf.nn.softmax(tf.transpose(prod, perm=[0, 2, 1]))
output = tf.concat([
code_output,
tf.matmul(prob, output[:, :, None])[:, :, 0],
tf.matmul(prob2, code_output[:, :, None])[:, :, 0]
],
axis=-1)
output = relu(fully_connected(output, 1024))
output = dropout(output, 0.6)
output = relu(fully_connected(output, 512))
output = dropout(output, 0.6)
output = relu(fully_connected(output, 256))
output = dropout(output, 0.8)
output = fully_connected(output, 5)
return output
