def NN(self, x):
assert x.get_shape().as_list()[:3] == [None, 101, 101]
summary_images(x, "layer0")
x = nn.convolution(x, 16, w=4) # 98
x = nn.convolution(x) # 96
summary_images(x, "layer2")
x = nn.max_pool(x)
x = nn.batch_normalization(x, self.tfacc)
########################################################################
assert x.get_shape().as_list() == [None, 48, 48, 16]
x = nn.convolution(x, 32) # 46
x = nn.convolution(x) # 44
summary_images(x, "layer4")
x = nn.max_pool(x)
x = nn.batch_normalization(x, self.tfacc)
########################################################################
assert x.get_shape().as_list() == [None, 22, 22, 32]
x = nn.convolution(x, 64) # 20
x = nn.convolution(x) # 18
summary_images(x, "layer6")
x = nn.max_pool(x)
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
########################################################################
assert x.get_shape().as_list() == [None, 9, 9, 64]
x = nn.convolution(x, 128) # 7
summary_images(x, "layer7")
x = tf.nn.dropout(x, self.tfkp)
x = nn.convolution(x) # 5
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
########################################################################
assert x.get_shape().as_list() == [None, 5, 5, 128]
x = nn.convolution(x, 1024, w=5)
########################################################################
assert x.get_shape().as_list() == [None, 1, 1, 1024]
x = tf.reshape(x, [-1, x.get_shape().as_list()[-1]])
self.embedding_input = x
x = tf.nn.dropout(x, self.tfkp)
x = nn.fullyconnected(x, 1024)
x = tf.nn.dropout(x, self.tfkp)
x = nn.fullyconnected(x, 1024)
x = nn.batch_normalization(x, self.tfacc)
self.test = x
x = nn.fullyconnected(x, 1, activation=None)
return x
def NN1(self, x):
assert x.get_shape().as_list()[:3] == [None, 101, 101]
x = nn.convolution(x, 16, w=4) # 98
x = nn.convolution(x) # 96
x = nn.max_pool(x)
x = nn.batch_normalization(x, self.tfacc)
########################################################################
assert x.get_shape().as_list() == [None, 48, 48, 16]
x = nn.convolution(x, 32) # 46
x = nn.convolution(x) # 44
x = nn.max_pool(x)
x = nn.batch_normalization(x, self.tfacc)
########################################################################
assert x.get_shape().as_list() == [None, 22, 22, 32]
x = nn.convolution(x, 64) # 20
x = nn.convolution(x) # 18
x = nn.max_pool(x)
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
########################################################################
assert x.get_shape().as_list() == [None, 9, 9, 64]
x = nn.convolution(x, 128) # 7
x = tf.nn.dropout(x, self.tfkp)
x = nn.convolution(x) # 5
summary_images(x, "nn1")
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
########################################################################
assert x.get_shape().as_list() == [None, 5, 5, 128]
x = nn.convolution(x, 1024, w=5)
########################################################################
assert x.get_shape().as_list() == [None, 1, 1, 1024]
x = tf.reshape(x, [-1, x.get_shape().as_list()[-1]])
x = tf.nn.dropout(x, self.tfkp)
x = nn.fullyconnected(x, 1024)
x = tf.nn.dropout(x, self.tfkp)
x = nn.fullyconnected(x, 1024)
x = nn.batch_normalization(x, self.tfacc)
self.test = x
x = nn.fullyconnected(x, 2, activation=None)
return x
def NN(self, x):
assert x.get_shape().as_list()[:3] == [None, 101, 101]
summary_images(x, "layer0")
x = nn.convolution(x, 16, w=2) # 100
summary_images(x, "layer1")
########################################################################
assert x.get_shape().as_list() == [None, 100, 100, 16]
x = nn.batch_normalization(x, self.tfacc)
x = res_layer(x, n=2) # 96
x = nn.batch_normalization(x, self.tfacc)
x = res_layer(x, n=2) # 92
summary_images(x, "layer5")
x = nn.max_pool(x)
########################################################################
assert x.get_shape().as_list() == [None, 46, 46, 16]
x = nn.batch_normalization(x, self.tfacc)
x = res_layer(x, 32, n=3) # 40
summary_images(x, "layer8")
x = nn.batch_normalization(x, self.tfacc)
x = res_layer(x, n=2) # 36
summary_images(x, "layer10")
x = nn.max_pool(x)
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
########################################################################
assert x.get_shape().as_list() == [None, 18, 18, 32]
x = res_layer(x, 64, n=2) # 14
summary_images(x, "layer12")
x = nn.batch_normalization(x, self.tfacc)
x = res_layer(x, 92, n=2) # 10
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
x = res_layer(x, 128, n=3) # 4
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
########################################################################
assert x.get_shape().as_list() == [None, 4, 4, 128]
x = nn.convolution(x, 1024, w=4)
########################################################################
assert x.get_shape().as_list() == [None, 1, 1, 1024]
x = tf.reshape(x, [-1, x.get_shape().as_list()[-1]])
self.embedding_input = x
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
x = nn.fullyconnected(x, 1024)
x = nn.batch_normalization(x, self.tfacc)
x = tf.nn.dropout(x, self.tfkp)
x = nn.fullyconnected(x, 1024)
x = nn.batch_normalization(x, self.tfacc)
self.test = x
x = nn.fullyconnected(x, 1, activation=None)
return x