ResNet significantly changed the view of how to parametrize the functions in deep networks. DenseNet (dense convolutional network) is to some extent the logical extension of this :cite:Huang.Liu.Van-Der-Maaten.ea.2017
.
To understand how to arrive at it, let us take a small detour to mathematics.
Recall the Taylor expansion for functions. For the point
The key point is that it decomposes a function into increasingly higher order terms. In a similar vein, ResNet decomposes functions into
That is, ResNet decomposes Huang.Liu.Van-Der-Maaten.ea.2017
.
As shown in :numref:fig_densenet_block
, the key difference between ResNet and DenseNet is that in the latter case outputs are concatenated (denoted by
In the end, all these functions are combined in MLP to reduce the number of features again. In terms of implementation this is quite simple:
rather than adding terms, we concatenate them. The name DenseNet arises from the fact that the dependency graph between variables becomes quite dense. The last layer of such a chain is densely connected to all previous layers. The dense connections are shown in :numref:fig_densenet
.
The main components that compose a DenseNet are dense blocks and transition layers. The former define how the inputs and outputs are concatenated, while the latter control the number of channels so that it is not too large.
DenseNet uses the modified "batch normalization, activation, and convolution"
structure of ResNet (see the exercise in :numref:sec_resnet
).
First, we implement this convolution block structure.
from d2l import mxnet as d2l
from mxnet import np, npx
from mxnet.gluon import nn
npx.set_np()
def conv_block(num_channels):
blk = nn.Sequential()
blk.add(nn.BatchNorm(),
nn.Activation('relu'),
nn.Conv2D(num_channels, kernel_size=3, padding=1))
return blk
#@tab pytorch
from d2l import torch as d2l
import torch
from torch import nn
def conv_block(input_channels, num_channels):
return nn.Sequential(
nn.BatchNorm2d(input_channels), nn.ReLU(),
nn.Conv2d(input_channels, num_channels, kernel_size=3, padding=1))
#@tab tensorflow
from d2l import tensorflow as d2l
import tensorflow as tf
class ConvBlock(tf.keras.layers.Layer):
def __init__(self, num_channels):
super(ConvBlock, self).__init__()
self.bn = tf.keras.layers.BatchNormalization()
self.relu = tf.keras.layers.ReLU()
self.conv = tf.keras.layers.Conv2D(
filters=num_channels, kernel_size=(3, 3), padding='same')
self.listLayers = [self.bn, self.relu, self.conv]
def call(self, x):
y = x
for layer in self.listLayers.layers:
y = layer(y)
y = tf.keras.layers.concatenate([x,y], axis=-1)
return y
A dense block consists of multiple convolution blocks, each using the same number of output channels. In the forward propagation, however, we concatenate the input and output of each convolution block on the channel dimension.
class DenseBlock(nn.Block):
def __init__(self, num_convs, num_channels, **kwargs):
super().__init__(**kwargs)
self.net = nn.Sequential()
for _ in range(num_convs):
self.net.add(conv_block(num_channels))
def forward(self, X):
for blk in self.net:
Y = blk(X)
# Concatenate the input and output of each block on the channel
# dimension
X = np.concatenate((X, Y), axis=1)
return X
#@tab pytorch
class DenseBlock(nn.Module):
def __init__(self, num_convs, input_channels, num_channels):
super(DenseBlock, self).__init__()
layer = []
for i in range(num_convs):
layer.append(conv_block(
num_channels * i + input_channels, num_channels))
self.net = nn.Sequential(*layer)
def forward(self, X):
for blk in self.net:
Y = blk(X)
# Concatenate the input and output of each block on the channel
# dimension
X = torch.cat((X, Y), dim=1)
return X
#@tab tensorflow
class DenseBlock(tf.keras.layers.Layer):
def __init__(self, num_convs, num_channels):
super(DenseBlock, self).__init__()
self.listLayers = []
for _ in range(num_convs):
self.listLayers.append(ConvBlock(num_channels))
def call(self, x):
for layer in self.listLayers.layers:
x = layer(x)
return x
In the following example,
we [define a DenseBlock
instance] with 2 convolution blocks of 10 output channels.
When using an input with 3 channels, we will get an output with
blk = DenseBlock(2, 10)
blk.initialize()
X = np.random.uniform(size=(4, 3, 8, 8))
Y = blk(X)
Y.shape
#@tab pytorch
blk = DenseBlock(2, 3, 10)
X = torch.randn(4, 3, 8, 8)
Y = blk(X)
Y.shape
#@tab tensorflow
blk = DenseBlock(2, 10)
X = tf.random.uniform((4, 8, 8, 3))
Y = blk(X)
Y.shape
Since each dense block will increase the number of channels, adding too many of them will lead to an excessively complex model. A transition layer is used to control the complexity of the model. It reduces the number of channels by using the
def transition_block(num_channels):
blk = nn.Sequential()
blk.add(nn.BatchNorm(), nn.Activation('relu'),
nn.Conv2D(num_channels, kernel_size=1),
nn.AvgPool2D(pool_size=2, strides=2))
return blk
#@tab pytorch
def transition_block(input_channels, num_channels):
return nn.Sequential(
nn.BatchNorm2d(input_channels), nn.ReLU(),
nn.Conv2d(input_channels, num_channels, kernel_size=1),
nn.AvgPool2d(kernel_size=2, stride=2))
#@tab tensorflow
class TransitionBlock(tf.keras.layers.Layer):
def __init__(self, num_channels, **kwargs):
super(TransitionBlock, self).__init__(**kwargs)
self.batch_norm = tf.keras.layers.BatchNormalization()
self.relu = tf.keras.layers.ReLU()
self.conv = tf.keras.layers.Conv2D(num_channels, kernel_size=1)
self.avg_pool = tf.keras.layers.AvgPool2D(pool_size=2, strides=2)
def call(self, x):
x = self.batch_norm(x)
x = self.relu(x)
x = self.conv(x)
return self.avg_pool(x)
[Apply a transition layer] with 10 channels to the output of the dense block in the previous example. This reduces the number of output channels to 10, and halves the height and width.
blk = transition_block(10)
blk.initialize()
blk(Y).shape
#@tab pytorch
blk = transition_block(23, 10)
blk(Y).shape
#@tab tensorflow
blk = TransitionBlock(10)
blk(Y).shape
Next, we will construct a DenseNet model. DenseNet first uses the same single convolutional layer and maximum pooling layer as in ResNet.
net = nn.Sequential()
net.add(nn.Conv2D(64, kernel_size=7, strides=2, padding=3),
nn.BatchNorm(), nn.Activation('relu'),
nn.MaxPool2D(pool_size=3, strides=2, padding=1))
#@tab pytorch
b1 = nn.Sequential(
nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
nn.BatchNorm2d(64), nn.ReLU(),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
#@tab tensorflow
def block_1():
return tf.keras.Sequential([
tf.keras.layers.Conv2D(64, kernel_size=7, strides=2, padding='same'),
tf.keras.layers.BatchNormalization(),
tf.keras.layers.ReLU(),
tf.keras.layers.MaxPool2D(pool_size=3, strides=2, padding='same')])
Then, similar to the four modules made up of residual blocks that ResNet uses,
DenseNet uses four dense blocks.
Similar to ResNet, we can set the number of convolutional layers used in each dense block. Here, we set it to 4, consistent with the ResNet-18 model in :numref:sec_resnet
. Furthermore, we set the number of channels (i.e., growth rate) for the convolutional layers in the dense block to 32, so 128 channels will be added to each dense block.
In ResNet, the height and width are reduced between each module by a residual block with a stride of 2. Here, we use the transition layer to halve the height and width and halve the number of channels.
# `num_channels`: the current number of channels
num_channels, growth_rate = 64, 32
num_convs_in_dense_blocks = [4, 4, 4, 4]
for i, num_convs in enumerate(num_convs_in_dense_blocks):
net.add(DenseBlock(num_convs, growth_rate))
# This is the number of output channels in the previous dense block
num_channels += num_convs * growth_rate
# A transition layer that halves the number of channels is added between
# the dense blocks
if i != len(num_convs_in_dense_blocks) - 1:
num_channels //= 2
net.add(transition_block(num_channels))
#@tab pytorch
# `num_channels`: the current number of channels
num_channels, growth_rate = 64, 32
num_convs_in_dense_blocks = [4, 4, 4, 4]
blks = []
for i, num_convs in enumerate(num_convs_in_dense_blocks):
blks.append(DenseBlock(num_convs, num_channels, growth_rate))
# This is the number of output channels in the previous dense block
num_channels += num_convs * growth_rate
# A transition layer that halves the number of channels is added between
# the dense blocks
if i != len(num_convs_in_dense_blocks) - 1:
blks.append(transition_block(num_channels, num_channels // 2))
num_channels = num_channels // 2
#@tab tensorflow
def block_2():
net = block_1()
# `num_channels`: the current number of channels
num_channels, growth_rate = 64, 32
num_convs_in_dense_blocks = [4, 4, 4, 4]
for i, num_convs in enumerate(num_convs_in_dense_blocks):
net.add(DenseBlock(num_convs, growth_rate))
# This is the number of output channels in the previous dense block
num_channels += num_convs * growth_rate
# A transition layer that halves the number of channels is added
# between the dense blocks
if i != len(num_convs_in_dense_blocks) - 1:
num_channels //= 2
net.add(TransitionBlock(num_channels))
return net
Similar to ResNet, a global pooling layer and a fully-connected layer are connected at the end to produce the output.
net.add(nn.BatchNorm(),
nn.Activation('relu'),
nn.GlobalAvgPool2D(),
nn.Dense(10))
#@tab pytorch
net = nn.Sequential(
b1, *blks,
nn.BatchNorm2d(num_channels), nn.ReLU(),
nn.AdaptiveMaxPool2d((1, 1)),
nn.Flatten(),
nn.Linear(num_channels, 10))
#@tab tensorflow
def net():
net = block_2()
net.add(tf.keras.layers.BatchNormalization())
net.add(tf.keras.layers.ReLU())
net.add(tf.keras.layers.GlobalAvgPool2D())
net.add(tf.keras.layers.Flatten())
net.add(tf.keras.layers.Dense(10))
return net
Since we are using a deeper network here, in this section, we will reduce the input height and width from 224 to 96 to simplify the computation.
#@tab all
lr, num_epochs, batch_size = 0.1, 10, 256
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=96)
d2l.train_ch6(net, train_iter, test_iter, num_epochs, lr, d2l.try_gpu())
- In terms of cross-layer connections, unlike ResNet, where inputs and outputs are added together, DenseNet concatenates inputs and outputs on the channel dimension.
- The main components that compose DenseNet are dense blocks and transition layers.
- We need to keep the dimensionality under control when composing the network by adding transition layers that shrink the number of channels again.
- Why do we use average pooling rather than maximum pooling in the transition layer?
- One of the advantages mentioned in the DenseNet paper is that its model parameters are smaller than those of ResNet. Why is this the case?
- One problem for which DenseNet has been criticized is its high memory consumption.
- Is this really the case? Try to change the input shape to
$224\times 224$ to see the actual GPU memory consumption. - Can you think of an alternative means of reducing the memory consumption? How would you need to change the framework?
- Is this really the case? Try to change the input shape to
- Implement the various DenseNet versions presented in Table 1 of the DenseNet paper :cite:
Huang.Liu.Van-Der-Maaten.ea.2017
. - Design an MLP-based model by applying the DenseNet idea. Apply it to the housing price prediction task in :numref:
sec_kaggle_house
.
:begin_tab:mxnet
Discussions
:end_tab:
:begin_tab:pytorch
Discussions
:end_tab:
:begin_tab:tensorflow
Discussions
:end_tab: