torchvision里densenet代码分析

#densenet原文地址 https://arxiv.org/abs/1608.06993 
#densenet介绍 https://blog.csdn.net/zchang81/article/details/76155291
#以下代码就是densenet在torchvision.models里的源码，为了提高自身的模型构建能力尝试分析下源代码：
import re
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.model_zoo as model_zoo
from collections import OrderedDict

__all__ = ['DenseNet', 'densenet121', 'densenet169', 'densenet201', 'densenet161']

model_urls = {
    'densenet121': 'https://download.pytorch.org/models/densenet121-a639ec97.pth',
    'densenet169': 'https://download.pytorch.org/models/densenet169-b2777c0a.pth',
    'densenet201': 'https://download.pytorch.org/models/densenet201-c1103571.pth',
    'densenet161': 'https://download.pytorch.org/models/densenet161-8d451a50.pth',
}   #这个是预训练模型可以在下边的densenet121,169等里直接在pretrained=True就可以下载 

def densenet121(pretrained=False, **kwargs):  #这是densenet121 返回一个在ImageNet上的预训练模型  #
    r"""Densenet-121 model from
    `"Densely Connected Convolutional Networks" <https://arxiv.org/pdf/1608.06993.pdf>`_

    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = DenseNet(num_init_features=64, growth_rate=32, block_config=(6, 12, 24, 16),
                     **kwargs)         #这里是模型的主要构建，使用了DenseNet类 直接就看Densenet类#
    if pretrained:
        # '.'s are no longer allowed in module names, but pervious _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
#把densenet169等就删除了，和上边的结构相同。 #

class DenseNet(nn.Module):  #这就是densenet的主类了，看继承了nn.Modele类 #
    r"""Densenet-BC model class, based on
    `"Densely Connected Convolutional Networks" <https://arxiv.org/pdf/1608.06993.pdf>`_

    Args:
        growth_rate (int) - how many filters to add each layer (`k` in paper) #每个denseblock里应该，每个Layer的输出特征数，就是论文里的k #
        block_config (list of 4 ints) - how many layers in each pooling block  #每个denseblock里layer层数, block_config的长度表示block的个数 #
        num_init_features (int) - the number of filters to learn in the first convolution layer   #初始化层里卷积输出的channel数#
        bn_size (int) - multiplicative factor for number of bottle neck layers   #这个是在block里一个denselayer里两个卷积层间的channel数 需要bn_size*growth_rate  #
          (i.e. bn_size * k features in the bottleneck layer)
        drop_rate (float) - dropout rate after each dense layer  #dropout的概率，正则化的方法 #
        num_classes (int) - number of classification classes   #输出的类别数，看后边接的是linear，应该最后加损失函数的时候应该加softmax，或者交叉熵，而且是要带计算概率的函数 #
    """
    def __init__(self, growth_rate=32, block_config=(6, 12, 24, 16),
                 num_init_features=64, bn_size=4, drop_rate=0, num_classes=1000):

        super(DenseNet, self).__init__()

        # First convolution  #初始化层，图像进来后不是直接进入denseblock，先使用大的卷积核，大步长，进一步压缩图像尺寸 #
　　　　 # 注意的是nn.Sequential的用法，ordereddict使用的方法，给layer命名，还有就是各层的排列，conv->bn->relu->pool 经过这一个操作就是尺寸就成为了1/4，数据量压缩了#
        self.features = nn.Sequential(OrderedDict([
            ('conv0', nn.Conv2d(3, num_init_features, kernel_size=7, stride=2, padding=3, bias=False)),
            ('norm0', nn.BatchNorm2d(num_init_features)),
            ('relu0', nn.ReLU(inplace=True)),
            ('pool0', nn.MaxPool2d(kernel_size=3, stride=2, padding=1)),
        ]))    #这里使用了batchnorm2d batchnorm 最近有group norm 是否可以换 #

        # Each denseblock 创建denseblock
        num_features = num_init_features
        for i, num_layers in enumerate(block_config):  #根据block_config里关于每个denseblock里的layer数量产生响应的block #
            block = _DenseBlock(num_layers=num_layers, num_input_features=num_features,
                                bn_size=bn_size, growth_rate=growth_rate, drop_rate=drop_rate)   #这是产生一个denseblock #
            self.features.add_module('denseblock%d' % (i + 1), block)   #加入到 nn.Sequential 里 #
            num_features = num_features + num_layers * growth_rate     #每一个denseblock最后输出的channel，因为是dense连接所以原始的输出有，也有内部每一层的特征 #
            if i != len(block_config) - 1:  #如果不是最后一层 #
                trans = _Transition(num_input_features=num_features, num_output_features=num_features // 2)   #transition层是压缩输出的特征数量为一半#
                self.features.add_module('transition%d' % (i + 1), trans)
                num_features = num_features // 2

        # Final batch norm
        self.features.add_module('norm5', nn.BatchNorm2d(num_features))

        # Linear layer
        self.classifier = nn.Linear(num_features, num_classes)

        # Official init from torch repo.
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal(m.weight.data)
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()
            elif isinstance(m, nn.Linear):
                m.bias.data.zero_()

    def forward(self, x):
        features = self.features(x)
        out = F.relu(features, inplace=True)
        out = F.avg_pool2d(out, kernel_size=7, stride=1).view(features.size(0), -1)
        out = self.classifier(out)
        return out

class _DenseLayer(nn.Sequential):    #这是denselayer，也是nn.Seqquential，看来要好好学习用法 #
    def __init__(self, num_input_features, growth_rate, bn_size, drop_rate):
        super(_DenseLayer, self).__init__()
        self.add_module('norm1', nn.BatchNorm2d(num_input_features)),               #这里要看到denselayer里其实主要包括两个卷积层，而且他们的channel数值得关注 #
        self.add_module('relu1', nn.ReLU(inplace=True)),                            #其实在add_module后边的逗号可以去掉，没有任何意义，又不是构成元组徒增歧义 #
        self.add_module('conv1', nn.Conv2d(num_input_features, 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)),            #这里注意的是输出的channel数是growth_rate #
        self.drop_rate = drop_rate

    def forward(self, x):            #这里是前传，主要解决的就是要把输出整形，把layer的输出和输入要cat在一起 #
        new_features = super(_DenseLayer, self).forward(x)   # #
        if self.drop_rate > 0:
            new_features = F.dropout(new_features, p=self.drop_rate, training=self.training)     #加入dropout增加泛化 #
        return torch.cat([x, new_features], 1)   #在channel上cat在一起，以形成dense连接 #

class _DenseBlock(nn.Sequential):   #是nn.Sequential的子类，将一个block里的layer组合起来 #
    def __init__(self, num_layers, num_input_features, bn_size, growth_rate, drop_rate):
        super(_DenseBlock, self).__init__()
        for i in range(num_layers):
            layer = _DenseLayer(num_input_features + i * growth_rate, growth_rate, bn_size, drop_rate)    #后一层的输入channel是该denseblock的输入channel数，加上该层前面层的channnel数的和 #
            self.add_module('denselayer%d' % (i + 1), layer)        

class _Transition(nn.Sequential):   #是nn.Sequential的子类，#这个就比较容易了，也是以后自己搭建module的案例#
    def __init__(self, num_input_features, num_output_features):          
        super(_Transition, self).__init__()
        self.add_module('norm', nn.BatchNorm2d(num_input_features))       
        self.add_module('relu', nn.ReLU(inplace=True))
        self.add_module('conv', nn.Conv2d(num_input_features, num_output_features,
                                          kernel_size=1, stride=1, bias=False))
        self.add_module('pool', nn.AvgPool2d(kernel_size=2, stride=2))'pool', nn.AvgPool2d(kernel_size=2, stride=2))

大概就是这样，作为去年最好的分类框架densenet，里边有很多学习的地方。

可以给自己搭建网络提供参考。