def attach_multibox_head(network, source_layer_names,
num_priors=4, num_classes=10, activation='softmax'):
heads = []
for idx, layer_name in enumerate(source_layer_names):
source_layer = network.get_layer(layer_name).output
# Classification
clf = Conv2D(num_priors * num_classes, (3, 3),
padding='same', name=f'clf_head{idx}_logit')(source_layer)
clf = Reshape((-1, num_classes),
name=f'clf_head{idx}_reshape')(clf)
if activation == 'softmax':
clf = Softmax(axis=-1, name=f'clf_head{idx}')(clf)
elif activation == 'sigmoid':
clf = sigmoid(clf)
else:
raise ValueError('activation은 {softmax,sigmoid}중에서 되어야 합니다.')
# Localization
loc = Conv2D(num_priors * 4, (3,3), padding='same',
name=f'loc_head{idx}')(source_layer)
loc = Reshape((-1,4),
name=f'loc_head{idx}_reshape')(loc)
head = Concatenate(axis=-1, name=f'head{idx}')([clf, loc])
heads.append(head)
if len(heads) > 1:
predictions = Concatenate(axis=1, name='predictions')(heads)
else:
predictions = K.identity(heads[0],name='predictions')
return predictions
def call(self, x, sigmoid_activation=True):
x = self.conv1(x)
x = self.conv2(x)
x = self.flatten(x)
x = self.fully_connected1(x)
x = self.fully_connected2(x)
if sigmoid_activation:
return sigmoid(x)
return relu(x)
def attach_multibox_head(network,
source_layer_names,
num_priors=4,
num_classes=11,
activation='softmax'): # 10->11
heads = []
batch_size = 64 # OpenCV에서 인식을못해서 하드코딩한다 reshape를 인식못한다
for idx, layer_name in enumerate(source_layer_names):
source_layer = network.get_layer(layer_name).output
# OpenCV Loading Error
# "Can't create layer \"loc_head2_reshape_2/Shape\" of type \"Shape\""
# 조치 : class개수를 10에서 11로 바꿔줌
w = source_layer.get_shape().as_list()[1]
h = source_layer.get_shape().as_list()[2]
print("w : ", w)
print("h : ", h)
print("num_priors : ", num_priors)
# Classification
clf = Conv2D(num_priors * num_classes, (3, 3),
padding='same',
name=f'clf_head{idx}_logit')(source_layer)
print("clf shape입니다 : ", clf.shape)
clf = tf.reshape(clf,
shape=(batch_size, w * h * num_priors, num_classes),
name=f'clf_head{idx}_reshape')
# clf = Reshape((w*h*num_priors, num_classes), name=f'clf_head{idx}_reshape')(clf) # (-1, num_classes) # w*h*num_priors
print("clf의 reshape 후입니다 : ", clf.shape)
if activation == 'softmax':
clf = Softmax(axis=-1, name=f'clf_head{idx}')(clf)
elif activation == 'sigmoid':
clf = sigmoid(clf)
else:
raise ValueError('activation은 {softmax,sigmoid}중에서 되어야 합니다.')
# Localization
loc = Conv2D(num_priors * 4, (3, 3),
padding='same',
name=f'loc_head{idx}')(source_layer)
print("loc의 shape입니다 : ", loc.shape)
loc = tf.reshape(loc,
shape=(batch_size, w * h * num_priors, 4),
name='loc_head{}_reshape'.format(idx))
# loc = Reshape((w*h*num_priors, 4), name=f'loc_head{idx}_reshape')(loc) #Reshape((-1, 4),
print("loc의 reshape 후입니다 : ", loc.shape)
head = Concatenate(axis=-1, name=f'head{idx}')([clf, loc])
heads.append(head)
if len(heads) > 1:
predictions = Concatenate(axis=1, name='predictions')(heads)
else:
predictions = K.identity(heads[0], name='predictions')
return predictions
def test_sigmoid(self):
def ref_sigmoid(x):
if x >= 0:
return 1 / (1 + np.exp(-x))
else:
z = np.exp(x)
return z / (1 + z)
sigmoid = np.vectorize(ref_sigmoid)
x = backend.placeholder(ndim=2)
f = backend.function([x], [activations.sigmoid(x)])
test_values = np.random.random((2, 5))
result = f([test_values])[0]
expected = sigmoid(test_values)
self.assertAllClose(result, expected, rtol=1e-05)
def make_generator(name, s, adj, node_f, use_gcn=True, use_gru=True):
n = node_f.shape[0] # number of nodes
input_s = Input(shape=(s, n))
input_f = Input(shape=(n, node_f.shape[1]))
input_g = Input(shape=(n, n))
if use_gcn:
gcov1 = GraphConv(2 * base)([input_f, input_g])
# gcov2 = GraphConv(base)([gcov1, input_g])
input_s1 = Dot(axes=(2, 1))([
input_s, gcov1
]) # dot product: element by element multiply, (1,s,n).(1,n,n)=(1,s,n)
else:
input_s1 = input_s
fc1 = Dense(4 * base, activation='relu', input_shape=(n, ))(input_s1)
fc2 = Dense(8 * base, activation='relu', input_shape=(n, ))(fc1)
# S*D2
if use_gru:
rnn1 = Dropout(dropout)(CuDNNGRU(2 * base, return_sequences=True)(fc2))
else:
rnn1 = fc2
fc3 = Dense(16 * base, activation='relu', input_shape=(n, ))(rnn1)
out = sigmoid(Dense(1)(fc3))
return Model(name=name, inputs=[input_s, input_f, input_g], outputs=out)
def call(self, inputs):
x = relu(self.linear1(inputs))
x = relu(self.linear2(x))
x = relu(self.linear3(x))
x = sigmoid(self.linear4(x))
return x