【PyTorch】SSD基础网络层的源码分析

SSD网络结构

多尺度stage （基础VGG+额外5个stage）

             

        

PyTorch源码及注释 

 源码：https://github.com/amdegroot/ssd.pytorch

# vgg([64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'C', 512, 512, 512, 'M',
#             512, 512, 512], 3)
def vgg(cfg, i, batch_norm=False):
    layers = [] #vgg网络的基本结构
    in_channels = i
    for v in cfg:  #遍历cfg列表
        if v == 'M':
            layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
        elif v == 'C':#ceil_mode :True，计算输出信号大小的时候，会使用向上取整，代替默认的向下取的操作
            layers += [nn.MaxPool2d(kernel_size=2, stride=2, ceil_mode=True)]
        else:
            conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=1)
            if batch_norm: #归一化之后在进行池化
                layers += [conv2d, nn.BatchNorm2d(v), nn.ReLU(inplace=True)]
            else:
                layers += [conv2d, nn.ReLU(inplace=True)]
            in_channels = v #输出层作为下一层输入
    pool5 = nn.MaxPool2d(kernel_size=3, stride=1, padding=1)
    conv6 = nn.Conv2d(512, 1024, kernel_size=3, padding=6, dilation=6)
    conv7 = nn.Conv2d(1024, 1024, kernel_size=1)
    layers += [pool5, conv6,
               nn.ReLU(inplace=True), conv7, nn.ReLU(inplace=True)]
    return layers

# add_extras([256, 'S', 512, 128, 'S', 256, 128, 256, 128, 256], 1024)
def add_extras(cfg, i, batch_norm=False):
    # Extra layers added to VGG for feature scaling（多尺度提取）
    layers = [] #具体功能需要看cfg的内容
    in_channels = i   #i=1024
    flag = False
    for k, v in enumerate(cfg):
        if in_channels != 'S':#跳过s的下一层
            if v == 'S':
                layers += [nn.Conv2d(in_channels, cfg[k + 1],
                           kernel_size=(1, 3)[flag], stride=2, padding=1)]
            else:#False=0,True表示1
                layers += [nn.Conv2d(in_channels, v, kernel_size=(1, 3)[flag])]
            flag = not flag
        in_channels = v
    return layers

# 基础vgg网络跟额外的stage层，cfg表示每一个stage边框的个数
# multibox(vgg, extra_layers, [4, 6, 6, 6, 4, 4], 21)
def multibox(vgg, extra_layers, cfg, num_classes):
    loc_layers = []
    conf_layers = []
    vgg_source = [21, -2]  #元素为21，-2的list
    for k, v in enumerate(vgg_source): #0,21  1,-2
        loc_layers += [nn.Conv2d(vgg[v].out_channels,#每个边框四个位置信息
                                 cfg[k] * 4, kernel_size=3, padding=1)]
        conf_layers += [nn.Conv2d(vgg[v].out_channels,#每个边框分为num_classes 21类
                        cfg[k] * num_classes, kernel_size=3, padding=1)]
    for k, v in enumerate(extra_layers[1::2], 2): #从索引1开始间隔2，遍历到最后
        loc_layers += [nn.Conv2d(v.out_channels, cfg[k]
                                 * 4, kernel_size=3, padding=1)]
        conf_layers += [nn.Conv2d(v.out_channels, cfg[k]
                                  * num_classes, kernel_size=3, padding=1)]
    return vgg, extra_layers, (loc_layers, conf_layers)

base = {
    '300': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'C', 512, 512, 512, 'M',
            512, 512, 512],
    '512': [],
}
extras = {
    '300': [256, 'S', 512, 128, 'S', 256, 128, 256, 128, 256],
    '512': [],
}
mbox = {
    '300': [4, 6, 6, 6, 4, 4],  # number of boxes per feature map location
    '512': [],
}

#train.py中调用方式
# ssd_net = build_ssd('train', cfg['min_dim'], cfg['num_classes'])
def build_ssd(phase, size=300, num_classes=21):
    if phase != "test" and phase != "train":
        print("ERROR: Phase: " + phase + " not recognized")
        return
    if size != 300: #repr函数表示一种机器阅读的方式。
        print("ERROR: You specified size " + repr(size) + ". However, " +
              "currently only SSD300 (size=300) is supported!")
        return
    base_, extras_, head_ = multibox(vgg(base[str(size)], 3),
                                     add_extras(extras[str(size)], 1024),
                                     mbox[str(size)], num_classes)
    #将参数送入到SSD的初始化构造函数
    return SSD(phase, size, base_, extras_, head_, num_classes)

参考

 https://blog.csdn.net/h__ang/article/details/90316220

https://blog.csdn.net/qq_39571318/article/details/89788687