def _build_model(input_shape):
x = Input(shape=input_shape, name="input")
img_input = x
x = VggPreprocess()(x)
# Block 1
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='block1_conv1')(x)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='block1_conv2')(x)
x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)
# Block 2
x = Conv2D(128, (3, 3),
activation='relu',
padding='same',
name='block2_conv1')(x)
x = Conv2D(128, (3, 3),
activation='relu',
padding='same',
name='block2_conv2')(x)
x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)
# Block 3
x = Conv2D(256, (3, 3),
activation='relu',
padding='same',
name='block3_conv1')(x)
model = Model(img_input, x, name='vgg19')
return model
def _build_model(input_shape):
x = Input(shape=input_shape, name="input")
img_input = x
x = VggPreprocess()(x)
base_model = VGG19(input_shape=input_shape, include_top=False)
model = Model(inputs=base_model.input, outputs=base_model.get_layer(output_layer).output)
fname = os.path.join(MODEL_ROOT, "h5/vgg19_weights_tf_dim_ordering_tf_kernels_notop.h5")
model.load_weights(fname, by_name=True)
x = model(x)
feat_model = Model(img_input, x, name='vgg19_extractor')
return feat_model
def _build_model(input_shape):
x = Input(shape=input_shape, name="input")
img_input = x
x = VggPreprocess()(x)
# Block 1
x = Conv2D(32, (3, 3), strides=2, use_bias=False, padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu")(x)
# x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv2')(x)
x = DepthwiseConv2D((3, 3), use_bias=False, padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = Conv2D(64, (1, 1), use_bias=False, padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu")(x)
# x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)
# Block 2
# x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv1')(x)
x = DepthwiseConv2D((3, 3), strides=2, use_bias=False,
padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = Conv2D(128, (1, 1), use_bias=False, padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu")(x)
# x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv2')(x)
x = DepthwiseConv2D((3, 3), use_bias=False, padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = Conv2D(128, (1, 1), use_bias=False, padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu")(x)
# x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)
# Block 3
# x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv1')(x)
x = DepthwiseConv2D((3, 3), use_bias=False, padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = Conv2D(256, (1, 1), use_bias=False, padding='same')(x)
if USE_TF_KERAS:
x = BatchNormalization(fused=False)(x)
else:
x = BatchNormalization()(x)
x = Activation("relu", name="block3_conv1")(x)
model = Model(img_input, x, name='mobile_encoder_31')
return model
def mobile_encoder(input_size=256, h5_fname=MOBILE_ENCODER_H5):
input_shape=[input_size,input_size,3]
x = Input(shape=input_shape, name="input")
img_input = x
# Block 1
x = VggPreprocess()(x)
x = Conv2D(32, (3, 3), strides=2, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (112,112,32)
x = DepthwiseConv2D((3, 3), use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
x = Conv2D(64, (1, 1), use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (112,112,64)
x = DepthwiseConv2D((3, 3), strides=2, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (56,56,64)
x = Conv2D(128, (1, 1), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (56,56,128)
x = DepthwiseConv2D((3, 3), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
x = Conv2D(128, (1, 1), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (56,56,128)
x = DepthwiseConv2D((3, 3), strides=2, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (28,28,128)
x = Conv2D(256, (1, 1), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (28,28,256)
x = DepthwiseConv2D((3, 3), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
x = Conv2D(256, (1, 1), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (28,28,256)
x = DepthwiseConv2D((3, 3), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
x = Conv2D(512, (1, 1), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
# (28,28,512)
x = DepthwiseConv2D((3, 3), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu")(x)
x = Conv2D(512, (1, 1), strides=1, use_bias=False, padding='same')(x)
x = BatchNormalization(fused=False)(x)
x = Activateion("relu", name="output")(x)
model = Model(img_input, x, name='vgg19_light')
model.load_weights(h5_fname)
return model