'''DenseNet models for Keras.
# Reference
- [Densely Connected Convolutional Networks](https://arxiv.org/pdf/1608.06993.pdf)
- [The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation](https://arxiv.org/pdf/1611.09326.pdf)
'''
from __future__ import print_function
from __future__ import absolute_import
from __future__ import division
import warnings
from tensorflow.python.keras.models import Model
from tensorflow.python.keras.layers import Dense, Dropout, Activation, Reshape
from tensorflow.python.keras.layers import Conv2D, Conv2DTranspose, UpSampling2D
from tensorflow.python.keras.layers import AveragePooling2D, MaxPooling2D
from tensorflow.python.keras.layers import GlobalAveragePooling2D
from tensorflow.python.keras.layers import Input
from tensorflow.python.keras.layers import concatenate
from tensorflow.python.keras.layers import BatchNormalization
from tensorflow.python.keras.regularizers import l2
from tensorflow.python.keras.utils import convert_all_kernels_in_model
from ..DenseNet.imagenet_utils import _obtain_input_shape
from ..DenseNet.imagenet_utils import decode_predictions
from tensorflow.python.keras import backend as K
from ..DenseNet.subpixel import SubPixelUpscaling
def preprocess_input(x, data_format=None):
"""Preprocesses a tensor encoding a batch of images.
# Arguments
x: input Numpy tensor, 4D.
data_format: data format of the image tensor.
# Returns
Preprocessed tensor.
"""
if data_format is None:
data_format = K.image_data_format()
assert data_format in {'channels_last', 'channels_first'}
if data_format == 'channels_first':
if x.ndim == 3:
# 'RGB'->'BGR'
x = x[::-1, ...]
# Zero-center by mean pixel
x[0, :, :] -= 103.939
x[1, :, :] -= 116.779
x[2, :, :] -= 123.68
else:
x = x[:, ::-1, ...]
x[:, 0, :, :] -= 103.939
x[:, 1, :, :] -= 116.779
x[:, 2, :, :] -= 123.68
else:
# 'RGB'->'BGR'
x = x[..., ::-1]
# Zero-center by mean pixel
x[..., 0] -= 103.939
x[..., 1] -= 116.779
x[..., 2] -= 123.68
x *= 0.017 # scale values
return x
def DenseNet(model_input, depth=40, nb_dense_block=3, growth_rate=12, nb_filter=-1, nb_layers_per_block=-1,
bottleneck=False, reduction=0.0, dropout_rate=0.0, weight_decay=1e-4, subsample_initial_block=False,
include_top=True, weights=None, input_tensor=None,
classes=10, activation='softmax', model_path = '', initial_num_classes=None, transfer_with_full_training = True):
if weights == 'imagenet' and include_top and classes != 1000:
raise ValueError('If using `weights` as ImageNet with `include_top`'
' as true, `classes` should be 1000')
if activation not in ['softmax', 'sigmoid']:
raise ValueError('activation must be one of "softmax" or "sigmoid"')
if activation == 'sigmoid' and classes != 1:
raise ValueError('sigmoid activation can only be used when classes = 1')
# Determine proper input shape
img_input = model_input
x = __create_dense_net(classes, img_input, include_top, depth, nb_dense_block,
growth_rate, nb_filter, nb_layers_per_block, bottleneck, reduction,
dropout_rate, weight_decay, subsample_initial_block, activation, initial_num_classes=initial_num_classes)
# Ensure that the model takes into account
# any potential predecessors of `input_tensor`.
inputs = img_input
# Create model.
model = Model(inputs, x, name='densenet')
# load weights
if (weights == 'imagenet'):
weights_path = model_path
model.load_weights(weights_path)
return model
elif (weights == "trained"):
weights_path = model_path
model.load_weights(weights_path)
return model
elif (weights == "continued"):
weights_path = model_path
model.load_weights(weights_path)
return model
elif (weights == "transfer"):
weights_path = model_path
model.load_weights(weights_path)
if (transfer_with_full_training == False):
for eachlayer in model.layers:
eachlayer.trainable = False
print("Training with top layers of the Model")
else:
print("Training with all layers of the Model")
x2 = model.layers[-2].output
x2 = Dense(classes, activation=activation)(x2)
new_model = Model(inputs=model.input, outputs=x2)
return new_model
elif (weights == "custom"):
return model
def DenseNetImageNet121(model_input=None,
bottleneck=True,
reduction=0.5,
dropout_rate=0.0,
weight_decay=1e-4,
include_top=True,
weights='imagenet',
input_tensor=None,
classes=1000,
activation='softmax', model_path = '', initial_num_classes= None, transfer_with_full_training = True):
return DenseNet(model_input, depth=121, nb_dense_block=4, growth_rate=32, nb_filter=64,
nb_layers_per_block=[6, 12, 24, 16], bottleneck=bottleneck, reduction=reduction,
dropout_rate=dropout_rate, weight_decay=weight_decay, subsample_initial_block=True,
include_top=include_top, weights=weights, input_tensor=input_tensor,
classes=classes, activation=activation, model_path=model_path, initial_num_classes=initial_num_classes, transfer_with_full_training = transfer_with_full_training)
def __conv_block(ip, nb_filter, bottleneck=False, dropout_rate=None, weight_decay=1e-4):
''' Apply BatchNorm, Relu, 3x3 Conv2D, optional bottleneck block and dropout
Args:
ip: Input keras tensor
nb_filter: number of filters
bottleneck: add bottleneck block
dropout_rate: dropout rate
weight_decay: weight decay factor
Returns: keras tensor with batch_norm, relu and convolution2d added (optional bottleneck)
'''
concat_axis = 1 if K.image_data_format() == 'channels_first' else -1
x = BatchNormalization(axis=concat_axis, epsilon=1.1e-5)(ip)
x = Activation('relu')(x)
if bottleneck:
inter_channel = nb_filter * 4 # Obtained from https://github.com/liuzhuang13/DenseNet/blob/master/densenet.lua
x = Conv2D(inter_channel, (1, 1), kernel_initializer='he_normal', padding='same', use_bias=False,
kernel_regularizer=l2(weight_decay))(x)
x = BatchNormalization(axis=concat_axis, epsilon=1.1e-5)(x)
x = Activation('relu')(x)
x = Conv2D(nb_filter, (3, 3), kernel_initializer='he_normal', padding='same', use_bias=False)(x)
if dropout_rate:
x = Dropout(dropout_rate)(x)
return x
def __dense_block(x, nb_layers, nb_filter, growth_rate, bottleneck=False, dropout_rate=None, weight_decay=1e-4,
grow_nb_filters=True, return_concat_list=False):
''' Build a dense_block where the output of each conv_block is fed to subsequent ones
Args:
x: keras tensor
nb_layers: the number of layers of conv_block to append to the model.
nb_filter: number of filters
growth_rate: growth rate
bottleneck: bottleneck block
dropout_rate: dropout rate
weight_decay: weight decay factor
grow_nb_filters: flag to decide to allow number of filters to grow
return_concat_list: return the list of feature maps along with the actual output
Returns: keras tensor with nb_layers of conv_block appended
'''
concat_axis = 1 if K.image_data_format() == 'channels_first' else -1
x_list = [x]
for i in range(nb_layers):
cb = __conv_block(x, growth_rate, bottleneck, dropout_rate, weight_decay)
x_list.append(cb)
x = concatenate([x, cb], axis=concat_axis)
if grow_nb_filters:
nb_filter += growth_rate
if return_concat_list:
return x, nb_filter, x_list
else:
return x, nb_filter
def __transition_block(ip, nb_filter, compression=1.0, weight_decay=1e-4):
''' Apply BatchNorm, Relu 1x1, Conv2D, optional compression, dropout and Maxpooling2D
Args:
ip: keras tensor
nb_filter: number of filters
compression: calculated as 1 - reduction. Reduces the number of feature maps
in the transition block.
dropout_rate: dropout rate
weight_decay: weight decay factor
Returns: keras tensor, after applying batch_norm, relu-conv, dropout, maxpool
'''
concat_axis = 1 if K.image_data_format() == 'channels_first' else -1
x = BatchNormalization(axis=concat_axis, epsilon=1.1e-5)(ip)
x = Activation('relu')(x)
x = Conv2D(int(nb_filter * compression), (1, 1), kernel_initializer='he_normal', padding='same', use_bias=False,
kernel_regularizer=l2(weight_decay))(x)
x = AveragePooling2D((2, 2), strides=(2, 2))(x)
return x
def __transition_up_block(ip, nb_filters, type='deconv', weight_decay=1E-4):
''' SubpixelConvolutional Upscaling (factor = 2)
Args:
ip: keras tensor
nb_filters: number of layers
type: can be 'upsampling', 'subpixel', 'deconv'. Determines type of upsampling performed
weight_decay: weight decay factor
Returns: keras tensor, after applying upsampling operation.
'''
if type == 'upsampling':
x = UpSampling2D()(ip)
elif type == 'subpixel':
x = Conv2D(nb_filters, (3, 3), activation='relu', padding='same', kernel_regularizer=l2(weight_decay),
use_bias=False, kernel_initializer='he_normal')(ip)
x = SubPixelUpscaling(scale_factor=2)(x)
x = Conv2D(nb_filters, (3, 3), activation='relu', padding='same', kernel_regularizer=l2(weight_decay),
use_bias=False, kernel_initializer='he_normal')(x)
else:
x = Conv2DTranspose(nb_filters, (3, 3), activation='relu', padding='same', strides=(2, 2),
kernel_initializer='he_normal', kernel_regularizer=l2(weight_decay))(ip)
return x
def __create_dense_net(nb_classes, img_input, include_top, depth=40, nb_dense_block=3, growth_rate=12, nb_filter=-1,
nb_layers_per_block=-1, bottleneck=False, reduction=0.0, dropout_rate=None, weight_decay=1e-4,
subsample_initial_block=False, activation='softmax', initial_num_classes=None):
''' Build the DenseNet model
Args:
nb_classes: number of classes
img_input: tuple of shape (channels, rows, columns) or (rows, columns, channels)
include_top: flag to include the final Dense layer
depth: number or layers
nb_dense_block: number of dense blocks to add to end (generally = 3)
growth_rate: number of filters to add per dense block
nb_filter: initial number of filters. Default -1 indicates initial number of filters is 2 * growth_rate
nb_layers_per_block: number of layers in each dense block.
Can be a -1, positive integer or a list.
If -1, calculates nb_layer_per_block from the depth of the network.
If positive integer, a set number of layers per dense block.
If list, nb_layer is used as provided. Note that list size must
be (nb_dense_block + 1)
bottleneck: add bottleneck blocks
reduction: reduction factor of transition blocks. Note : reduction value is inverted to compute compression
dropout_rate: dropout rate
weight_decay: weight decay rate
subsample_initial_block: Set to True to subsample the initial convolution and
add a MaxPool2D before the dense blocks are added.
subsample_initial:
activation: Type of activation at the top layer. Can be one of 'softmax' or 'sigmoid'.
Note that if sigmoid is used, classes must be 1.
Returns: keras tensor with nb_layers of conv_block appended
'''
concat_axis = 1 if K.image_data_format() == 'channels_first' else -1
if reduction != 0.0:
assert reduction <= 1.0 and reduction > 0.0, 'reduction value must lie between 0.0 and 1.0'
# layers in each dense block
if type(nb_layers_per_block) is list or type(nb_layers_per_block) is tuple:
nb_layers = list(nb_layers_per_block) # Convert tuple to list
assert len(nb_layers) == (nb_dense_block), 'If list, nb_layer is used as provided. ' \
'Note that list size must be (nb_dense_block)'
final_nb_layer = nb_layers[-1]
nb_layers = nb_layers[:-1]
else:
if nb_layers_per_block == -1:
assert (depth - 4) % 3 == 0, 'Depth must be 3 N + 4 if nb_layers_per_block == -1'
count = int((depth - 4) / 3)
if bottleneck:
count = count // 2
nb_layers = [count for _ in range(nb_dense_block)]
final_nb_layer = count
else:
final_nb_layer = nb_layers_per_block
nb_layers = [nb_layers_per_block] * nb_dense_block
# compute initial nb_filter if -1, else accept users initial nb_filter
if nb_filter <= 0:
nb_filter = 2 * growth_rate
# compute compression factor
compression = 1.0 - reduction
# Initial convolution
if subsample_initial_block:
initial_kernel = (7, 7)
initial_strides = (2, 2)
else:
initial_kernel = (3, 3)
initial_strides = (1, 1)
x = Conv2D(nb_filter, initial_kernel, kernel_initializer='he_normal', padding='same',
strides=initial_strides, use_bias=False, kernel_regularizer=l2(weight_decay))(img_input)
if subsample_initial_block:
x = BatchNormalization(axis=concat_axis, epsilon=1.1e-5)(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
# Add dense blocks
for block_idx in range(nb_dense_block - 1):
x, nb_filter = __dense_block(x, nb_layers[block_idx], nb_filter, growth_rate, bottleneck=bottleneck,
dropout_rate=dropout_rate, weight_decay=weight_decay)
# add transition_block
x = __transition_block(x, nb_filter, compression=compression, weight_decay=weight_decay)
nb_filter = int(nb_filter * compression)
# The last dense_block does not have a transition_block
x, nb_filter = __dense_block(x, final_nb_layer, nb_filter, growth_rate, bottleneck=bottleneck,
dropout_rate=dropout_rate, weight_decay=weight_decay)
x = BatchNormalization(axis=concat_axis, epsilon=1.1e-5)(x)
x = Activation('relu')(x)
x = GlobalAveragePooling2D(name="global_avg_pooling")(x)
if include_top:
if (initial_num_classes != None):
x = Dense(initial_num_classes, activation=activation)(x)
else:
x = Dense(nb_classes, activation=activation)(x)
return x
if __name__ == '__main__':
model = DenseNet((32, 32, 3), depth=100, nb_dense_block=3,
growth_rate=12, bottleneck=True, reduction=0.5, weights=None)
model.summary()
