def model(input_shape, n_classes):
in_net = Input(shape=input_shape, name='input')
net = Conv2d(64, (5, 5), (1, 1),
act=tf.nn.relu,
padding='SAME',
name='cnn1')(in_net)
net = MaxPool2d((3, 3), (2, 2), padding='SAME', name='pool1')(net)
net = LocalResponseNorm(4, 1.0, 0.001 / 9.0, 0.75, name='norm1')(net)
net = TernaryConv2d(64, (5, 5), (1, 1),
act=tf.nn.relu,
padding='SAME',
name='cnn2')(net)
net = LocalResponseNorm(4, 1.0, 0.001 / 9.0, 0.75, name='norm2')(net)
net = MaxPool2d((3, 3), (2, 2), padding='SAME', name='pool2')(net)
net = Flatten(name='flatten')(net)
net = TernaryDense(384, act=tf.nn.relu, name='d1relu')(net)
net = TernaryDense(192, act=tf.nn.relu, name='d2relu')(net)
net = Dense(n_classes, act=None, name='output')(net)
net = Model(inputs=in_net, outputs=net, name='dorefanet')
return net
def get_model(inputs_shape):
# self defined initialization
W_init = tl.initializers.truncated_normal(stddev=5e-2)
W_init2 = tl.initializers.truncated_normal(stddev=0.04)
b_init2 = tl.initializers.constant(value=0.1)
# build network
ni = Input(inputs_shape)
nn = Conv2d(64, (5, 5), (1, 1),
padding='SAME',
act=tf.nn.relu,
W_init=W_init,
b_init=None,
name='conv1')(ni)
nn = MaxPool2d((3, 3), (2, 2), padding='SAME', name='pool1')(nn)
nn = LocalResponseNorm(depth_radius=4,
bias=1.0,
alpha=0.001 / 9.0,
beta=0.75,
name="norm1")(nn)
nn = Conv2d(64, (5, 5), (1, 1),
padding='SAME',
act=tf.nn.relu,
W_init=W_init,
b_init=None,
name='conv2')(nn)
nn = LocalResponseNorm(depth_radius=4,
bias=1.0,
alpha=0.001 / 9.0,
beta=0.75,
name="norm2")(nn)
nn = MaxPool2d((3, 3), (2, 2), padding='SAME', name='pool2')(nn)
nn = Flatten(name='flatten')(nn)
nn = Dense(384,
act=tf.nn.relu,
W_init=W_init2,
b_init=b_init2,
name='dense1relu')(nn)
nn = Dense(192,
act=tf.nn.relu,
W_init=W_init2,
b_init=b_init2,
name='dense2relu')(nn)
nn = Dense(10, act=None, W_init=W_init2, name='output')(nn)
M = Model(inputs=ni, outputs=nn, name='cnn')
return M
def dorefanet_model(input_shape, n_classes):
in_net = Input(shape=input_shape, name='input')
net = Conv2d(32, (5, 5), (1, 1), act='relu', padding='SAME', name='conv1')(in_net)
net = MaxPool2d((3, 3), (2, 2), padding='SAME', name='pool1')(net)
net = LocalResponseNorm(4, 1.0, 0.001 / 9.0, 0.75, name='norm1')(net)
net = tl.layers.Sign("sign")(net)
net = DorefaConv2d(8, 32, 64, (5, 5), (1, 1), act='relu', padding='SAME', name='DorefaConv1')(net)
net = LocalResponseNorm(4, 1.0, 0.001 / 9.0, 0.75, name='norm2')(net)
net = MaxPool2d((3, 3), (2, 2), padding='SAME', name='pool2')(net)
net = Flatten(name='flatten')(net)
net = DorefaDense(8, 16, 384, act='relu', name='DorefaDense1')(net)
net = DorefaDense(8, 16, 192, act='relu', name='DorefaDense2')(net)
net = Dense(n_classes, act=None, name='output')(net)
net = Model(inputs=in_net, outputs=net, name='dorefanet')
return net
def binary_model(input_shape, n_classes):
in_net = Input(shape=input_shape, name='input')
net = Conv2d(64, (5, 5), (1, 1), act='relu', padding='SAME',
name='conv1')(in_net)
net = Sign(name='sign1')(net)
net = MaxPool2d((3, 3), (2, 2), padding='SAME', name='pool1')(net)
net = LocalResponseNorm(4, 1.0, 0.001 / 9.0, 0.75, name='norm1')(net)
net = BinaryConv2d(64, (5, 5), (1, 1),
act='relu',
padding='SAME',
name='bconv1')(net)
net = LocalResponseNorm(4, 1.0, 0.001 / 9.0, 0.75, name='norm2')(net)
net = MaxPool2d((3, 3), (2, 2), padding='SAME', name='pool2')(net)
net = Flatten(name='flatten')(net)
net = Sign(name='sign2')(net)
net = BinaryDense(384, act='relu', name='d1relu')(net)
net = Sign(name='sign3')(net)
net = BinaryDense(192, act='relu', name='d2relu')(net)
net = Dense(n_classes, act=None, name='output')(net)
net = Model(inputs=in_net, outputs=net, name='binarynet')
return net