def TruncatedMobileNet(input_height, input_width):
assert input_height // 16 * 16 == input_height
assert input_width // 16 * 16 == input_width
alpha = 1.0
depth_multiplier = 1
img_input = Input(shape=[input_height, input_width, 3], name='image_input')
# s / 2
x = _conv_block(img_input, 32, alpha, strides=(2, 2))
x_s2 = _depthwise_conv_block(x, 64, alpha, depth_multiplier, block_id=1)
# s / 4
x = _depthwise_conv_block(x_s2, 128, alpha, depth_multiplier,
strides=(2, 2), block_id=2)
x_s4 = _depthwise_conv_block(x, 128, alpha, depth_multiplier, block_id=3)
# s / 8
x = _depthwise_conv_block(x_s4, 256, alpha, depth_multiplier,
strides=(2, 2), block_id=4)
x_s8 = _depthwise_conv_block(x, 256, alpha, depth_multiplier, block_id=5)
# s / 16
# DONE(see--): Make some of these dilated
x = _depthwise_conv_block(x_s8, 512, alpha, depth_multiplier,
strides=(2, 2), block_id=6)
x = _depthwise_conv_block(
x, 512, alpha, depth_multiplier, dilation_rate=1, block_id=7)
x = _depthwise_conv_block(
x, 512, alpha, depth_multiplier, dilation_rate=1, block_id=8)
x = _depthwise_conv_block(
x, 512, alpha, depth_multiplier, dilation_rate=2, block_id=9)
x = _depthwise_conv_block(
x, 512, alpha, depth_multiplier, dilation_rate=4, block_id=10)
x_s16 = _depthwise_conv_block(
x, 512, alpha, depth_multiplier, dilation_rate=4, block_id=11)
return img_input, x_s2, x_s4, x_s8, x_s16
def My_MobileNet(triplet, alpha=1.0, depth_multiplier=1):
img_input = triplet
x = _conv_block(img_input, 32, alpha, strides=(2, 2))
x = _depthwise_conv_block(x, 64, alpha, depth_multiplier, block_id=1)
x = _depthwise_conv_block(x,
128,
alpha,
depth_multiplier,
strides=(2, 2),
block_id=2)
x = _depthwise_conv_block(x, 128, alpha, depth_multiplier, block_id=3)
x = _depthwise_conv_block(x,
256,
alpha,
depth_multiplier,
strides=(2, 2),
block_id=4)
x = _depthwise_conv_block(x, 256, alpha, depth_multiplier, block_id=5)
x = _depthwise_conv_block(x,
512,
alpha,
depth_multiplier,
strides=(2, 2),
block_id=6)
x = _depthwise_conv_block(x, 512, alpha, depth_multiplier, block_id=7)
x = _depthwise_conv_block(x, 512, alpha, depth_multiplier, block_id=8)
x = _depthwise_conv_block(x, 512, alpha, depth_multiplier, block_id=9)
x = _depthwise_conv_block(x, 512, alpha, depth_multiplier, block_id=10)
x = _depthwise_conv_block(x, 512, alpha, depth_multiplier, block_id=11)
x = _depthwise_conv_block(x,
1024,
alpha,
depth_multiplier,
strides=(2, 2),
block_id=12)
x = _depthwise_conv_block(x, 1024, alpha, depth_multiplier, block_id=13)
x = GlobalAveragePooling2D()(x)
x = Flatten()(x)
x = Dense(128)(x)
return x
def MobileNetSlim(input_shape, alpha, depth_multiplier=1, output_classes=1, dropout=0.4):
input = Input(shape=input_shape, name='flow')
x = _conv_block(input, 32, alpha, strides=(2, 2))
x = _depthwise_conv_block(x, 64, alpha, depth_multiplier, block_id=1)
x = _depthwise_conv_block(x, 128, alpha, depth_multiplier, strides=(2, 2), block_id=2)
x = _depthwise_conv_block(x, 128, alpha, depth_multiplier, block_id=3)
x = _depthwise_conv_block(x, 256, alpha, depth_multiplier, strides=(2, 2), block_id=4)
x1 = _depthwise_conv_block(x, 256, alpha, depth_multiplier, block_id=5)
x = GlobalMaxPooling2D()(x)
x = Dense(64, kernel_regularizer=regularizers.l2(0.01))(x)
x = Activation('elu')(x)
output = Dense(1, name='speed')(x)
model = Model(inputs=input, outputs=output, name='optical_flow_encoder')
return model
def mobilenet_block(x, weight_decay, palpha=1.0, depth_multiplier=1):
x = _conv_block(x, 32, palpha, strides=(2, 2))
x = _depthwise_conv_block_mod(x, 64, palpha, depth_multiplier, block_id=1)
x = _depthwise_conv_block_mod(x,
128,
palpha,
depth_multiplier,
strides=(2, 2),
block_id=2)
x = _depthwise_conv_block_mod(x, 128, palpha, depth_multiplier, block_id=3)
x = _depthwise_conv_block_mod(x,
256,
palpha,
depth_multiplier,
strides=(2, 2),
block_id=4)
x = _depthwise_conv_block_mod(x, 256, palpha, depth_multiplier, block_id=5)
# x = _depthwise_conv_block_mod(x, 512, palpha, depth_multiplier, strides=(2, 2), block_id=6)
x = _depthwise_conv_block_mod(x,
512,
palpha,
depth_multiplier,
strides=(1, 1),
block_id=6)
x = _depthwise_conv_block_mod(x, 512, palpha, depth_multiplier, block_id=7)
x = _depthwise_conv_block_mod(x, 512, palpha, depth_multiplier, block_id=8)
x = _depthwise_conv_block_mod(x, 512, palpha, depth_multiplier, block_id=9)
x = _depthwise_conv_block_mod(x,
512,
palpha,
depth_multiplier,
block_id=10)
# x = _depthwise_conv_block_mod(x, 512, palpha, depth_multiplier, block_id=11)
# x = _depthwise_conv_block_mod(x, 1024, palpha, depth_multiplier, strides=(2, 2), block_id=12)
# x = _depthwise_conv_block_mod(x, 1024, palpha, depth_multiplier, block_id=13)
return x