def create_efficientnetb0(height,
width,
pretrained: bool,
mode: XModelMode = XModelMode.SIMPLE):
shape = (height, width, 3)
efficientnet.layers.BatchNormalization = BatchNormalization
base_model = efficientnet.EfficientNetB0(input_shape=shape,
include_top=False,
weights=None)
if pretrained:
base_model.load_weights(load_weights("imagenet",
model_name="efficientnetb0"),
by_name=True,
skip_mismatch=True)
print("\033[31m", "Imagenet model loaded", "\033[0m")
else:
print("\033[31m", "Imagenet model not loaded", "\033[0m")
inputs = tf.keras.Input(shape=shape, name="input")
base_model, features = upsample(
base_model,
inputs,
[
"block2b_activation", "block3b_activation", "block5c_activation",
"top_activation"
], # 144, 240, 672, 1280
# ["block2b_add", "block3b_add", "block5c_add", "block6d_add"], # 24, 40, 112, 192
# resnet50: 256, 512, 1024, 2048
mode,
)
return base_model, inputs, features
def create_efficientnetb2(height, width, pretrained: bool, mode: XModelMode = XModelMode.SIMPLE):
shape = (height, width, 3)
tf.keras.layers.BatchNormalization = BatchNormalization
base_model = EfficientNetB2(input_shape=shape, include_top=False, weights="imagenet" if pretrained else None)
inputs = tf.keras.Input(shape=shape, name="input")
base_model, features = upsample(
base_model, inputs, ["block2c_add", "block3c_add", "block5d_add", "block6e_add"], mode
)
return base_model, inputs, features
def create_efficientnetb0(height, width, pretrained: bool, mode: XModelMode = XModelMode.SIMPLE):
shape = (height, width, 3)
tf.keras.layers.BatchNormalization = BatchNormalization
base_model = EfficientNetB0(input_shape=shape, include_top=False, weights="imagenet" if pretrained else None)
inputs = tf.keras.Input(shape=shape, name="input")
base_model, features = upsample(
base_model,
inputs,
["block2b_activation", "block3b_activation", "block5c_activation", "top_activation"], # 144, 240, 672, 1280
# ["block2b_add", "block3b_add", "block5c_add", "block6d_add"], # 24, 40, 112, 192
# resnet50: 256, 512, 1024, 2048
mode,
)
return base_model, inputs, features
def create_efficientnetb2(height,
width,
pretrained: bool,
mode: XModelMode = XModelMode.SIMPLE):
shape = (height, width, 3)
base_model = efficientnet.EfficientNetB2(input_shape=shape,
include_top=False,
weights=None)
if pretrained:
base_model.load_weights(load_weights("imagenet",
model_name="efficientnetb2"),
by_name=True,
skip_mismatch=True)
inputs = tf.keras.Input(shape=shape, name="input")
base_model, features = upsample(
base_model, inputs,
["block2c_add", "block3c_add", "block5d_add", "block6e_add"], mode)
return base_model, inputs, features
def create_resnet_18(height,
width,
pretrained: bool,
mode: XModelMode = XModelMode.SIMPLE):
shape = (height, width, 3)
ResNet18, preprocess_input = Classifiers.get("resnet18")
weights = "imagenet" if pretrained else None
base_model = ResNet18(input_shape=shape,
weights=weights,
include_top=False)
base_model = set_regularization(
base_model, kernel_regularizer=tf.keras.regularizers.l2(L2_REG))
inputs = tf.keras.Input(shape=shape, name="input")
base_model, features = upsample(base_model, inputs, [
"stage2_unit1_relu1", "stage3_unit1_relu1", "stage4_unit1_relu1",
"relu1"
], mode)
return base_model, inputs, features
def create_resnet(height,
width,
pretrained: bool,
mode: XModelMode = XModelMode.SIMPLE):
shape = (height, width, 3)
tf.keras.layers.BatchNormalization = BatchNormalization
base_model = tf.keras.applications.resnet50.ResNet50(
input_shape=shape,
include_top=False,
weights="imagenet" if pretrained else None,
layers=tf.keras.layers)
inputs = tf.keras.Input(shape=shape, name="input")
x = PreprocessCaffeLayer()(inputs)
base_model, features = upsample(base_model, x, [
"conv2_block3_out", "conv3_block3_out", "conv4_block3_out",
"conv5_block3_out"
], mode)
return base_model, inputs, features