def get_resnet_model(save_path, model_res=1024, image_size=256, depth=2, size=0, activation='elu', loss='logcosh', optimizer='adam'):
# Build model
if os.path.exists(save_path):
print('Loading model')
return load_model(save_path)
print('Building model')
model_scale = int(2*(math.log(model_res,2)-1)) # For example, 1024 -> 18
if size <= 0:
from keras.applications.resnet50 import ResNet50
resnet = ResNet50(include_top=False, pooling=None, weights='imagenet', input_shape=(image_size, image_size, 3))
else:
from keras_applications.resnet_v2 import ResNet50V2, ResNet101V2, ResNet152V2
if size == 1:
resnet = ResNet50V2(include_top=False, pooling=None, weights='imagenet', input_shape=(image_size, image_size, 3), backend = keras.backend, layers = keras.layers, models = keras.models, utils = keras.utils)
if size == 2:
resnet = ResNet101V2(include_top=False, pooling=None, weights='imagenet', input_shape=(image_size, image_size, 3), backend = keras.backend, layers = keras.layers, models = keras.models, utils = keras.utils)
if size >= 3:
resnet = ResNet152V2(include_top=False, pooling=None, weights='imagenet', input_shape=(image_size, image_size, 3), backend = keras.backend, layers = keras.layers, models = keras.models, utils = keras.utils)
layer_size = model_scale*8*8*8
if is_square(layer_size): # work out layer dimensions
layer_l = int(math.sqrt(layer_size)+0.5)
layer_r = layer_l
else:
layer_m = math.log(math.sqrt(layer_size),2)
layer_l = 2**math.ceil(layer_m)
layer_r = layer_size // layer_l
layer_l = int(layer_l)
layer_r = int(layer_r)
x_init = None
inp = Input(shape=(image_size, image_size, 3))
x = resnet(inp)
if (depth < 0):
depth = 1
if (size <= 1):
if (size <= 0):
x = Conv2D(model_scale*8, 1, activation=activation)(x) # scale down
x = Reshape((layer_r, layer_l))(x)
else:
x = Conv2D(model_scale*8*4, 1, activation=activation)(x) # scale down a little
x = Reshape((layer_r*2, layer_l*2))(x)
else:
if (size == 2):
x = Conv2D(1024, 1, activation=activation)(x) # scale down a bit
x = Reshape((256, 256))(x)
else:
x = Reshape((256, 512))(x) # all weights used
while (depth > 0): # See https://github.com/OliverRichter/TreeConnect/blob/master/cifar.py - TreeConnect inspired layers instead of dense layers.
x = LocallyConnected1D(layer_r, 1, activation=activation)(x)
x = Permute((2, 1))(x)
x = LocallyConnected1D(layer_l, 1, activation=activation)(x)
x = Permute((2, 1))(x)
if x_init is not None:
x = Add()([x, x_init]) # add skip connection
x_init = x
depth-=1
x = Reshape((model_scale, 512))(x) # train against all dlatent values
model = Model(inputs=inp,outputs=x)
model.compile(loss=loss, metrics=[], optimizer=optimizer) # By default: adam optimizer, logcosh used for loss.
return model
def load_backbone(backbone_type="resnet50",
backbone_outputs=('C3', 'C4', 'C5', 'P6', 'P7'),
num_features=256):
global BACKBONE_LAYERS
inputs = Input((None, None, 3), name='images')
if backbone_type.lower() == 'resnet50':
preprocess = BackBonePreProcess(rgb=False,
mean_shift=True,
normalize=0)(inputs)
model = ResNet50(input_tensor=preprocess, include_top=False)
elif backbone_type.lower() == 'resnet50v2':
preprocess = BackBonePreProcess(rgb=True, mean_shift=True,
normalize=2)(inputs)
resnet50v2, _ = Classifiers.get('resnet50v2')
model = resnet50v2(input_tensor=preprocess,
include_top=False,
weights='imagenet')
elif backbone_type.lower() == "resnet101v2":
preprocess = BackBonePreProcess(rgb=True,
mean_shift=False,
normalize=2)(inputs)
model = ResNet101V2(input_tensor=preprocess,
include_top=False,
backend=tf.keras.backend,
layers=tf.keras.layers,
models=tf.keras.models,
utils=tf.keras.utils)
elif backbone_type.lower() == 'resnext50':
preprocess = BackBonePreProcess(rgb=True, mean_shift=True,
normalize=2)(inputs)
model = ResNeXt50(input_tensor=preprocess, include_top=False)
elif backbone_type.lower() == "seresnet50":
preprocess = BackBonePreProcess(rgb=True, mean_shift=True,
normalize=3)(inputs)
seresnet50, _ = Classifiers.get('seresnet50')
model = seresnet50(input_tensor=preprocess,
original_input=inputs,
include_top=False,
weights='imagenet')
elif backbone_type.lower() == "seresnet34":
preprocess = BackBonePreProcess(rgb=True,
mean_shift=False,
normalize=0)(inputs)
seresnet34, _ = Classifiers.get('seresnet34')
model = seresnet34(input_tensor=preprocess,
original_input=inputs,
include_top=False,
weights='imagenet')
elif backbone_type.lower() == "seresnext50":
preprocess = BackBonePreProcess(rgb=True, mean_shift=True,
normalize=3)(inputs)
seresnext50, _ = Classifiers.get('seresnext50')
model = seresnext50(input_tensor=preprocess,
original_input=inputs,
include_top=False,
weights='imagenet')
elif backbone_type.lower() == "vgg16":
preprocess = BackBonePreProcess(rgb=False,
mean_shift=True,
normalize=0)(inputs)
model = VGG16(input_tensor=preprocess, include_top=False)
elif backbone_type.lower() == "mobilenet":
preprocess = BackBonePreProcess(rgb=False,
mean_shift=False,
normalize=2)(inputs)
model = MobileNet(input_tensor=preprocess,
include_top=False,
alpha=1.0)
elif backbone_type.lower() == 'efficientnetb2':
preprocess = BackBonePreProcess(rgb=True, mean_shift=True,
normalize=3)(inputs)
model = efn.EfficientNetB2(input_tensor=preprocess,
include_top=False,
weights='imagenet')
elif backbone_type.lower() == 'efficientnetb3':
preprocess = BackBonePreProcess(rgb=True, mean_shift=True,
normalize=3)(inputs)
model = efn.EfficientNetB3(input_tensor=preprocess,
include_top=False,
weights='imagenet')
elif backbone_type.lower() == 'efficientnetb4':
preprocess = BackBonePreProcess(rgb=True, mean_shift=True,
normalize=3)(inputs)
model = efn.EfficientNetB4(input_tensor=preprocess,
include_top=False,
weights='imagenet')
else:
raise NotImplementedError(
f"backbone_type은 {BACKBONE_LAYERS.keys()} 중에서 하나가 되어야 합니다.")
model.trainable = False
# Block Layer 가져오기
features = []
for key, layer_name in BACKBONE_LAYERS[backbone_type.lower()].items():
if key in backbone_outputs:
layer_tensor = model.get_layer(layer_name).output
features.append(Identity(name=key)(layer_tensor))
if backbone_type.lower() == "mobilenet":
# Extra Layer for Feature Extracting
Z6 = ZeroPadding2D(((0, 1), (0, 1)),
name=f'P6_zeropadding')(features[-1])
P6 = Conv2D(num_features, (3, 3),
strides=(2, 2),
padding='valid',
activation='relu',
name=f'P6_conv')(Z6)
if 'P6' in backbone_outputs:
features.append(Identity(name='P6')(P6))
G6 = GroupNormalization(name=f'P6_norm')(P6)
Z7 = ZeroPadding2D(((0, 1), (0, 1)), name=f'P7_zeropadding')(G6)
P7 = Conv2D(num_features, (3, 3),
strides=(2, 2),
padding='valid',
activation='relu',
name=f'P7_conv')(Z7)
if 'P7' in backbone_outputs:
features.append(Identity(name=f'P7')(P7))
else:
P6 = Conv2D(num_features, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
name=f'P6_conv')(features[-1])
if 'P6' in backbone_outputs:
features.append(Identity(name=f'P6')(P6))
G6 = GroupNormalization(name=f'P6_norm')(P6)
P7 = Conv2D(num_features, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
name=f'P7_conv')(G6)
if 'P7' in backbone_outputs:
features.append(Identity(name=f'P7')(P7))
return Model(inputs, features, name=backbone_type)