J3、DenseNet算法实战与解析

• 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
• 🍦 参考文章：365天深度学习训练营-第P2周：彩色识别
• 🍖 原作者：K同学啊|接辅导、项目定制

 

 

学习要求¶

• 根据本文 Pytorch 代码，编写出相应的 TensorFlow 代码（建议使用上周的数据测试一下模型是否构建正确）
• 了解并研究 DenseNet与ResNetV 的区别

学习提高¶

• 改进思路是否可以迁移到其他地方呢

 

 

ResNet模型的核心是通过建立前面层与后面层之间的“短路连接”（shortcuts，skip connection），进而训练出更深的CNN网络。

DenseNet模型的基本思路与ResNet一致，但是它建立的是前面所有层与后面层的密集连接（dense connection），它的名称也是由此而来。DenseNet的另一大特色是通过特征在channel上的连接来实现特征重用（feature reuse）。这些特点让DenseNet在参数和计算成本更少的情形下实现比ResNet更优的性能，DenseNet也因此斩获CVPR 2017的最佳论文奖。

论文原文：Densely Connected Convolutional Networks.pdf

 

 

一、论文解读¶

1、设计理念¶

相比ResNet，DenseNet提出了一个更激进的密集连接机制：即互相连接所有的层，具体来说就是每个层都会接受其前面所有层作为其额外的输入。

ResNet是每个层与前面的某层（一般是2-4层）短路连接在一起，连接方式是通过元素相加。

而在DenseNet中，每个层都会与前面所有层在channel维度上连接（concat）在一起（即元素叠加），并作为下一层的输入。所有的层的输入都来源于前面所有层在channel维度的concat。

2、网络结构¶

CNN网络一般要经过Pooling或者stride>1的Conv来降低特征图的大小，而DenseNet的密集连接方式需要特征图大小保持一致。为了解决这个问题，DenseNet网络中使用DenseBlock+Transition的结构，其中DenseBlock是包含很多层的模块，每个层的特征图大小相同，层与层之间采用密集连接方式。而Transition层是连接两个相邻的DenseBlock，并且通过Pooling使特征图大小降低。

DenseBlock示意如下，各个DenseBlock之间通过Transition层连接在一起

在DenseBlock中，各个层的特征图大小一致，可以在channel维度上连接。DenseBlock中的非线性组合函数H(·)的是 BN+ReLU+3x3 Conv 的结构，如下图所示。另外值得注意的一点是，与ResNet不同，所有DenseBlock中各个层卷积之后均输出k个特征图，即得到的特征图的channel数为k，或者说采用k个卷积核。k在DenseNet称为growth rate，这是一个超参数。一般情况下使用较小的k（比如12），就可以得到较佳的性能。假定输入层的特征图的channel数为k0，那么l层输入的channel数为k~0~+k~(1,2,...,l-1)~，因此随着层数增加，尽管k设定得较小，DenseBlock的输入会非常多，不过这是由于特征重用所造成的，每个层仅有k个特征是自己独有的。

由于后面层的输入会非常大，DenseBlock内部可以采用bottleneck层来减少计算量，主要是原有的结构中增加1x1 Conv，如图7所示，即BN+ReLU+1x1 Conv+BN+ReLU+3x3 Conv，称为DenseNet-B结构。其中1x1 Conv得到4k个特征图它起到的作用是降低特征数量，从而提升计算效率。

对于Transition层，，它主要是连接两个相邻的DenseBlock，并且降低特征图大小。Transition层包括一个1x1的卷积和2x2的AvgPooling，结构为BN+ReLU+1x1Conv+2x2AvgPooling。另外，Transition层可以起到压缩模型的作用。假定层的上接DenseBlock得到的特征图channels数为m，Transition层可以产生【θm】个特征（通过卷积层），其中 是压缩系数θ∈（0，1]（compression rate）。当 θ=1时，特征个数经过Transition层没有变化，即无压缩，而当压缩系数小于1时，这种结构称为DenseNet-C，文中使用θ=0.5。对于使用bottleneck层的DenseBlock结构和压缩系数小于1的Transition组合结构称为DenseNet-BC。

对于ImageNet数据集，图片输入大小为224×224，网络结构采用包含4个DenseBlock的DenseNet-BC，其首先是一个stride=2的7x7卷积层，然后是一个stride=2的3x3 MaxPooling层，后面才进入DenseBlock。ImageNet数据集所采用的网络配置如下表所示：

3、与其他算法的对比¶

 

 

二、实现DenseNet121¶

打卡要求是使用tensorflow，但是因为我并没有太多的相关代码经验，所以可能会有问题，还请见谅

1、搭建DenseBlock内部结构：BN+ReLU+1x1Conv+BN+ReLU+3x3Conv+dropout¶

 

In [2]:

# PyTorch搭建DenseBlock
import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
import matplotlib.pyplot as plot
import torch.nn.functional as F

class DenseLayer(nn.Sequential):
    def __init__(self, in_channel, growth_rate, bn_size, drop_rate):
        super(DenseLayer, self).__init__()
        self.add_module('norm1', nn.BatchNorm2d(in_channel))
        self.add_module('relu1', nn.ReLU(inplace=True))
        self.add_module('conv1', nn.Conv2d(in_channel, bn_size*growth_rate,
                                           kernel_size=1, stride=1, bias=False))
        self.add_module('norm2', nn.BatchNorm2d(bn_size*growth_rate))
        self.add_module('relu2', nn.ReLU(inplace=True))
        self.add_module('conv2', nn.Conv2d(bn_size*growth_rate, growth_rate,
                                           kernel_size=3, stride=1, padding=1, bias=False))
        self.drop_rate = drop_rate
    
    def forward(self, x):
        new_feature = super(DenseLayer, self).forward(x)
        if self.drop_rate>0:
            new_feature = F.dropout(new_feature, p=self.drop_rate, training=self.training)
        return torch.cat([x, new_feature], 1)

 

In [ ]:

# TensorFlow搭建DenseBlock
import tensorflow as tf
from tensorflow.keras import layers, models
from tensorflow import keras

class BottleNeck(keras.Model):
    def __init__(self, growth_rate, bn_size = 4, dropout = 0.3):
        super().__init__()
        self.bn1 = layers.BatchNormalization()
        self.relu = layers.Activation("relu"),
        self.conv1 = layers.Conv2D(filters=bn_size * growth_rate, kernel_size=(1, 1), 
                                   strides=1, padding='same')
        self.bn2 = layers.BatchNormalization()
        self.conv2 = layers.Conv2D(filters=growth_rate, kernel_size=(3, 3), 
                                   strides=1, padding='same')
        self.dropout = layers.Dropout(rate=dropout)

        self.listLayers = [
            self.bn1,
            self.relu,
            self.conv1,
            self.bn2,
            self.relu,
            self.conv2,
            self.dropout
            ]
        
    def call(self, x):
        tem = x
        for layer in self.listLayers.layers:
            x = layer(x)
        return layers.concatenate([tem, x], axis=-1)

 

 

2、实现DenseBlock¶

 

In [6]:

# PyTorch
''' DenseBlock '''
class DenseBlock(nn.Sequential):
    def __init__(self, num_layers, in_channel, bn_size, growth_rate, drop_rate):
        super(DenseBlock, self).__init__()
        for i in range(num_layers):
            layer = DenseLayer(in_channel+i*growth_rate, growth_rate, bn_size, drop_rate)
            self.add_module('denselayer%d'%(i+1,), layer)

 

In [ ]:

# TensorFlow
class DenseBlock(tf.keras.Model):
    def __init__(self, num_layer, growth_rate, bn_size = 4, dropout = 0.3, efficient=False):
        super().__init__()
        self.efficient = efficient
        self.listLayers = []
        if self.efficient:
            _x = tf.recompute_grad(BottleNeck(growth_rate, bn_size = bn_size, dropout = dropout))
        else:_x =BottleNeck(growth_rate, bn_size = bn_size, dropout = dropout)
        for _ in range(num_layer):
            self.listLayers.append(BottleNeck(growth_rate, bn_size = bn_size, dropout = dropout))
    
    def call(self, x):
        for layer in self.listLayers.layers:
            x = layer(x)
        return x

 

 

3、实现Transition：主要是一个卷积层和一个池化层¶

 

In [7]:

# PyTorch
''' Transition layer between two adjacent DenseBlock '''
class Transition(nn.Sequential):
    def __init__(self, in_channel, out_channel):
        super(Transition, self).__init__()
        self.add_module('norm', nn.BatchNorm2d(in_channel))
        self.add_module('relu', nn.ReLU(inplace=True))
        self.add_module('conv', nn.Conv2d(in_channel, out_channel,
                                          kernel_size=1, stride=1, bias=False))
        self.add_module('pool', nn.AvgPool2d(2, stride=2))

 

In [ ]:

# TensorFlow
class Transition(tf.keras.Model):
    def __init__(self, growth_rate):
        super().__init__()
        self.bn1 = layers.BatchNormalization()
        self.relu = layers.Activation('relu')
        self.conv1 = layers.Conv2D(filters = growth_rate, kernel_size=(1, 1),
                                   strides = 1, activation = 'relu', padding='same')
        self.pooling = layers.AveragePooling2D(pool_size=(2,2), strides = 2,padding='same')

        self.listLayers = [
            self.bn1,
            self.relu,
            self.conv1,
            self.pooling
        ]
    def call(self,x):
        for layer in self.listLayers.layers:
            x = layer(x)
        return x

 

 

4、实现DenseNet网络¶

 

In [9]:

# PyTorch
class DenseNet(nn.Module):
    def __init__(self, growth_rate=32, block_config=(6,12,24,16), init_channel=64, 
                 bn_size=4, compression_rate=0.5, drop_rate=0, num_classes=1000):     
        """
        :param growth_rate: (int) number of filters used in DenseLayer, `k` in the paper
        :param block_config: (list of 4 ints) number of layers in eatch DenseBlock
        :param init_channel: (int) number of filters in the first Conv2d
        :param bn_size: (int) the factor using in the bottleneck layer
        :param compression_rate: (float) the compression rate used in Transition Layer
        :param drop_rate: (float) the drop rate after each DenseLayer
        :param num_classes: (int) number of classes for classification
        """
        super(DenseNet, self).__init__()
        # first Conv2d
        self.features = nn.Sequential(OrderedDict([
            ('conv0', nn.Conv2d(3, init_channel, kernel_size=7, stride=2, padding=3, bias=False)),
            ('norm0', nn.BatchNorm2d(init_channel)),
            ('relu0', nn.ReLU(inplace=True)),
            ('pool0', nn.MaxPool2d(3, stride=2, padding=1))
        ]))
        # DenseBlock
        num_features = init_channel
        for i, num_layers in enumerate(block_config):
            block = DenseBlock(num_layers, num_features, bn_size, growth_rate, drop_rate)
            self.features.add_module('denseblock%d'%(i+1), block)
            num_features += num_layers*growth_rate
            if i != len(block_config)-1:
                transition = Transition(num_features, int(num_features*compression_rate))
                self.features.add_module('transition%d'%(i+1), transition)
                num_features = int(num_features*compression_rate)
        # final BN+ReLU
        self.features.add_module('norm5', nn.BatchNorm2d(num_features))
        self.features.add_module('relu5', nn.ReLU(inplace=True))
        # classification layer
        self.classifier = nn.Linear(num_features, num_classes)
        # params initialization
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.bias, 0)
                nn.init.constant_(m.weight, 1)
            elif isinstance(m, nn.Linear):
                nn.init.constant_(m.bias, 0)
    
    def forward(self, x):
        x = self.features(x)
        x = F.avg_pool2d(x, 7, stride=1).view(x.size(0), -1)
        x = self.classifier(x)
        return x

 

In [14]:

# TensorFlow
class DenseNet(tf.keras.Model):
    def __init__(self, num_init_feature, growth_rate, block_config, num_classes, 
                 bn_size=4, dropout=0.3, compression_rate=0.5, efficient=False):
        super().__init__()
        self.num_channels = num_init_feature
        self.conv = layers.Conv2D(filters = num_init_feature, kernel_size=7,
                                strides = 2, padding='same')
        self.bn = layers.BatchNormalization()
        self.relu = layers.Activation('relu')
        self.max_pool = layers.MaxPool2D(pool_size=3, strides=2, padding='same')

        self.dense_block_layers = []
        for i in block_config[:-1]:
            self.dense_block_layers.append( DenseBlock(num_layer =i, growth_rate = growth_rate,
                                                 bn_size = bn_size, dropout = dropout, efficient=efficient))
            self.num_channels = compression_rate * (self.num_channels + growth_rate * i)
            self.dense_block_layers.append( Transition(self.num_channels))
            
        self.dense_block_layers.append( DenseBlock(num_layer =block_config[-1], growth_rate = growth_rate,
                                             bn_size = bn_size, dropout = dropout, efficient=efficient))
        
        self.avgpool = layers.GlobalAveragePooling2D()
        self.fc = tf.keras.layers.Dense(units=num_classes, activation=tf.keras.activations.softmax)
            
    def call(self,x):
        x = self.conv(x)
        x = self.bn(x)
        x = self.relu(x)
        x = self.max_pool(x)
        
        for layer in self.dense_block_layers.layers:
            x = layer(x)

        
        x = self.avgpool(x)
        return self.fc(x)

 

 

5、实现DenseNet-121网络¶

 

In [12]:

# PtTorch
def densenet121(n_classes=1000, pretrained=False, **kwargs):
    model = DenseNet(init_channel=64, growth_rate=32, block_config=(6,12,24,16),
                     num_classes=n_classes, **kwargs)
    if pretrained:
        # '.'s are no longer allowed in module names, but previous DenseLayer
        # has keys 'norm.1', 'relu.1', 'conv.1', 'norm.2', 'relu.2', 'conv.2'.
        # They are also in the checkpoints in model_urls. This pattern is used
        # to find such keys.
        pattern = re.compile(
            r'^(.*denselayer\d+\.(?:norm|relu|conv))\.((?:[12])\.(?:weight|bias|running_mean|running_var))$')
        state_dict = model_zoo.load_url(model_urls['densenet121'])
        for key in list(state_dict.keys()):
            res = pattern.match(key)
            if res:
                new_key = res.group(1) + res.group(2)
                state_dict[new_key] = state_dict[key]
                del state_dict[key]
        model.load_state_dict(state_dict)
    return model

 

In [ ]:

# TensorFlow
model = DenseNet(num_init_feature=64,
                 growth_rate=32,
                 block_config=[6,12,24,16],
                 compression_rate=0.5,
                 num_classes = 4,
                 dropout=0.0,
                 efficient=True)

 

 

三、DenseNet121网络结构图¶