def identity_block(self, input, kernel_size, filters):
filters1, filters2, filters3 = filters
output = Conv2D(num_outputs=filters1,
kernel_size=1,
activation='linear',
padding='same',
in_layers=[input])
output = BatchNorm(in_layers=[output])
output = ReLU(output)
output = Conv2D(num_outputs=filters2,
kernel_size=kernel_size,
activation='linear',
padding='same',
in_layers=[input])
output = BatchNorm(in_layers=[output])
output = ReLU(output)
output = Conv2D(num_outputs=filters3,
kernel_size=1,
activation='linear',
padding='same',
in_layers=[input])
output = BatchNorm(in_layers=[output])
output = Add(in_layers=[output, input])
output = ReLU(output)
return output
def test_mnist(self):
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
train = dc.data.NumpyDataset(mnist.train.images, mnist.train.labels)
valid = dc.data.NumpyDataset(mnist.validation.images,
mnist.validation.labels)
# Images are square 28x28 (batch, height, width, channel)
feature = Feature(shape=(None, 784), name="Feature")
make_image = Reshape(shape=(-1, 28, 28, 1), in_layers=[feature])
conv2d_1 = Conv2D(num_outputs=32,
normalizer_fn=tf.contrib.layers.batch_norm,
in_layers=[make_image])
maxpool_1 = MaxPool(in_layers=[conv2d_1])
conv2d_2 = Conv2D(num_outputs=64,
normalizer_fn=tf.contrib.layers.batch_norm,
in_layers=[maxpool_1])
maxpool_2 = MaxPool(in_layers=[conv2d_2])
flatten = Flatten(in_layers=[maxpool_2])
dense1 = Dense(out_channels=1024,
activation_fn=tf.nn.relu,
in_layers=[flatten])
dense2 = Dense(out_channels=10, in_layers=[dense1])
label = Label(shape=(None, 10), name="Label")
smce = SoftMaxCrossEntropy(in_layers=[label, dense2])
loss = ReduceMean(in_layers=[smce])
output = SoftMax(in_layers=[dense2])
tg = dc.models.TensorGraph(model_dir='/tmp/mnist',
batch_size=1000,
use_queue=True)
tg.add_output(output)
tg.set_loss(loss)
tg.fit(train, nb_epoch=2)
prediction = np.squeeze(tg.predict_proba_on_batch(valid.X))
fpr = dict()
tpr = dict()
roc_auc = dict()
for i in range(10):
fpr[i], tpr[i], thresh = roc_curve(valid.y[:, i], prediction[:, i])
roc_auc[i] = auc(fpr[i], tpr[i])
assert_true(roc_auc[i] > 0.99)
def test_Conv2D_pickle(): tg = TensorGraph() feature = Feature(shape=(tg.batch_size, 10, 10, 1)) layer = Conv2D(num_outputs=3, in_layers=feature) tg.add_output(layer) tg.set_loss(layer) tg.build() tg.save()
def __init__(self,
img_rows=224,
img_cols=224,
weights="imagenet",
classes=1000,
**kwargs):
super(ResNet50, self).__init__(use_queue=False, **kwargs)
self.img_cols = img_cols
self.img_rows = img_rows
self.weights = weights
self.classes = classes
input = Feature(shape=(None, self.img_rows, self.img_cols, 3))
labels = Label(shape=(None, self.classes))
conv1 = Conv2D(num_outputs=64,
kernel_size=7,
stride=2,
activation='linear',
padding='same',
in_layers=[input])
bn1 = BatchNorm(in_layers=[conv1])
ac1 = ReLU(bn1)
pool1 = MaxPool2D(ksize=[1, 3, 3, 1], in_layers=[bn1])
cb1 = self.conv_block(pool1, 3, [64, 64, 256], 1)
id1 = self.identity_block(cb1, 3, [64, 64, 256])
id1 = self.identity_block(id1, 3, [64, 64, 256])
cb2 = self.conv_block(id1, 3, [128, 128, 512])
id2 = self.identity_block(cb2, 3, [128, 128, 512])
id2 = self.identity_block(id2, 3, [128, 128, 512])
id2 = self.identity_block(id2, 3, [128, 128, 512])
cb3 = self.conv_block(id2, 3, [256, 256, 1024])
id3 = self.identity_block(cb3, 3, [256, 256, 1024])
id3 = self.identity_block(id3, 3, [256, 256, 1024])
id3 = self.identity_block(id3, 3, [256, 256, 1024])
id3 = self.identity_block(cb3, 3, [256, 256, 1024])
id3 = self.identity_block(id3, 3, [256, 256, 1024])
cb4 = self.conv_block(id3, 3, [512, 512, 2048])
id4 = self.identity_block(cb4, 3, [512, 512, 2048])
id4 = self.identity_block(id4, 3, [512, 512, 2048])
pool2 = AvgPool2D(ksize=[1, 7, 7, 1], in_layers=[id4])
flatten = Flatten(in_layers=[pool2])
dense = Dense(classes, in_layers=[flatten])
loss = SoftMaxCrossEntropy(in_layers=[labels, dense])
loss = ReduceMean(in_layers=[loss])
self.set_loss(loss)
self.add_output(dense)
def conv_block(self, input, kernel_size, filters, strides=2):
filters1, filters2, filters3 = filters
output = Conv2D(num_outputs=filters1,
kernel_size=1,
stride=strides,
activation='linear',
padding='same',
in_layers=[input])
output = BatchNorm(in_layers=[output])
output = ReLU(output)
output = Conv2D(num_outputs=filters2,
kernel_size=kernel_size,
activation='linear',
padding='same',
in_layers=[output])
output = BatchNorm(in_layers=[output])
output = ReLU(output)
output = Conv2D(num_outputs=filters3,
kernel_size=1,
activation='linear',
padding='same',
in_layers=[output])
output = BatchNorm(in_layers=[output])
shortcut = Conv2D(num_outputs=filters3,
kernel_size=1,
stride=strides,
activation='linear',
padding='same',
in_layers=[input])
shortcut = BatchNorm(in_layers=[shortcut])
output = Add(in_layers=[shortcut, output])
output = ReLU(output)
return output
def test_conv_2D(self):
"""Test that Conv2D can be invoked."""
length = 4
width = 5
in_channels = 2
out_channels = 3
batch_size = 20
in_tensor = np.random.rand(batch_size, length, width, in_channels)
with self.test_session() as sess:
in_tensor = tf.convert_to_tensor(in_tensor, dtype=tf.float32)
out_tensor = Conv2D(out_channels, kernel_size=1)(in_tensor)
sess.run(tf.global_variables_initializer())
out_tensor = out_tensor.eval()
assert out_tensor.shape == (batch_size, length, width, out_channels)
def __init__(self,
img_rows=512,
img_cols=512,
filters=[64, 128, 256, 512, 1024],
model=dc.models.TensorGraph(),
**kwargs):
super(UNet, self).__init__(use_queue=False, **kwargs)
self.img_cols = img_cols
self.img_rows = img_rows
self.filters = filters
self.model = dc.models.TensorGraph()
input = Feature(shape=(None, self.img_rows, self.img_cols))
conv1 = Conv2D(num_outputs=self.filters[0],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[input])
conv1 = Conv2D(num_outputs=self.filters[0],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv1])
pool1 = MaxPool2D(ksize=2, in_layers=[conv1])
conv2 = Conv2D(num_outputs=self.filters[1],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[pool1])
conv2 = Conv2D(num_outputs=self.filters[1],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv2])
pool2 = MaxPool2D(ksize=2, in_layers=[conv2])
conv3 = Conv2D(num_outputs=self.filters[2],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[pool2])
conv3 = Conv2D(num_outputs=self.filters[2],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv3])
pool3 = MaxPool2D(ksize=2, in_layers=[conv3])
conv4 = Conv2D(num_outputs=self.filters[3],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[pool3])
conv4 = Conv2D(num_outputs=self.filters[3],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv4])
pool4 = MaxPool2D(ksize=2, in_layers=[conv4])
conv5 = Conv2D(num_outputs=self.filters[4],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[pool4])
conv5 = Conv2D(num_outputs=self.filters[4],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv5])
up6 = Conv2DTranspose(num_outputs=self.filters[3],
kernel_size=2,
in_layers=[conv5])
concat6 = Concat(in_layers=[conv4, up6], axis=1)
conv6 = Conv2D(num_outputs=self.filters[3],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[concat6])
conv6 = Conv2D(num_outputs=self.filters[3],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv6])
up7 = Conv2DTranspose(num_outputs=self.filters[2],
kernel_size=2,
in_layers=[conv6])
concat7 = Concat(in_layers=[conv3, up7], axis=1)
conv7 = Conv2D(num_outputs=self.filters[2],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[concat7])
conv7 = Conv2D(num_outputs=self.filters[2],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv7])
up8 = Conv2DTranspose(num_outputs=self.filters[1],
kernel_size=2,
in_layers=[conv7])
concat8 = Concat(in_layers=[conv2, up8], axis=1)
conv8 = Conv2D(num_outputs=self.filters[1],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[concat8])
conv8 = Conv2D(num_outputs=self.filters[1],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv8])
up9 = Conv2DTranspose(num_outputs=self.filters[0],
kernel_size=2,
in_layers=[conv8])
concat9 = Concat(in_layers=[conv1, up9], axis=1)
conv9 = Conv2D(num_outputs=self.filters[0],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[concat9])
conv9 = Conv2D(num_outputs=self.filters[0],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv9])
conv10 = Conv2D(num_outputs=1,
kernel_size=1,
activation='sigmoid',
in_layers=[conv9])
model.add_output(conv10)
def build_graph(self):
# inputs placeholder
self.inputs = Feature(shape=(None, self.image_size, self.image_size,
3),
dtype=tf.float32)
# data preprocessing and augmentation
in_layer = DRAugment(self.augment,
self.batch_size,
size=(self.image_size, self.image_size),
in_layers=[self.inputs])
# first conv layer
in_layer = Conv2D(int(self.n_init_kernel),
kernel_size=7,
activation_fn=None,
in_layers=[in_layer])
in_layer = BatchNorm(in_layers=[in_layer])
in_layer = ReLU(in_layers=[in_layer])
# downsample by max pooling
res_in = MaxPool2D(ksize=[1, 3, 3, 1],
strides=[1, 2, 2, 1],
in_layers=[in_layer])
for ct_module in range(self.n_downsample - 1):
# each module is a residual convolutional block
# followed by a convolutional downsample layer
in_layer = Conv2D(int(self.n_init_kernel * 2**(ct_module - 1)),
kernel_size=1,
activation_fn=None,
in_layers=[res_in])
in_layer = BatchNorm(in_layers=[in_layer])
in_layer = ReLU(in_layers=[in_layer])
in_layer = Conv2D(int(self.n_init_kernel * 2**(ct_module - 1)),
kernel_size=3,
activation_fn=None,
in_layers=[in_layer])
in_layer = BatchNorm(in_layers=[in_layer])
in_layer = ReLU(in_layers=[in_layer])
in_layer = Conv2D(int(self.n_init_kernel * 2**ct_module),
kernel_size=1,
activation_fn=None,
in_layers=[in_layer])
res_a = BatchNorm(in_layers=[in_layer])
res_out = res_in + res_a
res_in = Conv2D(int(self.n_init_kernel * 2**(ct_module + 1)),
kernel_size=3,
stride=2,
in_layers=[res_out])
res_in = BatchNorm(in_layers=[res_in])
# max pooling over the final outcome
in_layer = ReduceMax(axis=(1, 2), in_layers=[res_in])
for layer_size in self.n_fully_connected:
# fully connected layers
in_layer = Dense(layer_size,
activation_fn=tf.nn.relu,
in_layers=[in_layer])
# dropout for dense layers
#in_layer = Dropout(0.25, in_layers=[in_layer])
logit_pred = Dense(self.n_tasks * self.n_classes,
activation_fn=None,
in_layers=[in_layer])
logit_pred = Reshape(shape=(None, self.n_tasks, self.n_classes),
in_layers=[logit_pred])
weights = Weights(shape=(None, self.n_tasks))
labels = Label(shape=(None, self.n_tasks), dtype=tf.int32)
output = SoftMax(logit_pred)
self.add_output(output)
loss = SparseSoftMaxCrossEntropy(in_layers=[labels, logit_pred])
weighted_loss = WeightedError(in_layers=[loss, weights])
# weight decay regularizer
# weighted_loss = WeightDecay(0.1, 'l2', in_layers=[weighted_loss])
self.set_loss(weighted_loss)
def __init__(self,
img_rows=512,
img_cols=512,
filters=[64, 128, 256, 512, 1024],
**kwargs):
super(UNet, self).__init__(use_queue=False, **kwargs)
self.img_cols = img_cols
self.img_rows = img_rows
self.filters = filters
input = Feature(shape=(None, self.img_rows, self.img_cols, 3))
labels = Label(shape=(None, self.img_rows * self.img_cols))
conv1 = Conv2D(num_outputs=self.filters[0],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[input])
conv1 = Conv2D(num_outputs=self.filters[0],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv1])
pool1 = MaxPool2D(ksize=[1, 2, 2, 1], in_layers=[conv1])
conv2 = Conv2D(num_outputs=self.filters[1],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[pool1])
conv2 = Conv2D(num_outputs=self.filters[1],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv2])
pool2 = MaxPool2D(ksize=[1, 2, 2, 1], in_layers=[conv2])
conv3 = Conv2D(num_outputs=self.filters[2],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[pool2])
conv3 = Conv2D(num_outputs=self.filters[2],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv3])
pool3 = MaxPool2D(ksize=[1, 2, 2, 1], in_layers=[conv3])
conv4 = Conv2D(num_outputs=self.filters[3],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[pool3])
conv4 = Conv2D(num_outputs=self.filters[3],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv4])
pool4 = MaxPool2D(ksize=[1, 2, 2, 1], in_layers=[conv4])
conv5 = Conv2D(num_outputs=self.filters[4],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[pool4])
conv5 = Conv2D(num_outputs=self.filters[4],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv5])
up6 = Conv2DTranspose(num_outputs=self.filters[3],
kernel_size=2,
stride=2,
in_layers=[conv5])
concat6 = Concat(in_layers=[conv4, up6], axis=3)
conv6 = Conv2D(num_outputs=self.filters[3],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[concat6])
conv6 = Conv2D(num_outputs=self.filters[3],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv6])
up7 = Conv2DTranspose(num_outputs=self.filters[2],
kernel_size=2,
stride=2,
in_layers=[conv6])
concat7 = Concat(in_layers=[conv3, up7], axis=3)
conv7 = Conv2D(num_outputs=self.filters[2],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[concat7])
conv7 = Conv2D(num_outputs=self.filters[2],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv7])
up8 = Conv2DTranspose(num_outputs=self.filters[1],
kernel_size=2,
stride=2,
in_layers=[conv7])
concat8 = Concat(in_layers=[conv2, up8], axis=3)
conv8 = Conv2D(num_outputs=self.filters[1],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[concat8])
conv8 = Conv2D(num_outputs=self.filters[1],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv8])
up9 = Conv2DTranspose(num_outputs=self.filters[0],
kernel_size=2,
stride=2,
in_layers=[conv8])
concat9 = Concat(in_layers=[conv1, up9], axis=3)
conv9 = Conv2D(num_outputs=self.filters[0],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[concat9])
conv9 = Conv2D(num_outputs=self.filters[0],
kernel_size=3,
activation='relu',
padding='same',
in_layers=[conv9])
conv10 = Conv2D(num_outputs=1,
kernel_size=1,
activation='sigmoid',
in_layers=[conv9])
loss = SoftMaxCrossEntropy(in_layers=[labels, conv10])
loss = ReduceMean(in_layers=[loss])
self.set_loss(loss)
self.add_output(conv10)