HRNet源代码结构解析

一、整体结构

按顺序可以分为三个部分：

1、stem net

输入IMG，输出feature map，得到这个尺寸的特征图之后，HRNet始终保持此尺寸的图片

2、HRNet的四个阶段

（1）每个stage产生的multi-scale特征图，配置如表1。
（2）stage 的连接处有 transition 结构，用于在不同 stage 之间连接，完成 channels 及 feature map 大小对应。
表1

3、segment head

将stage4输出的4种scale特征concat到一起，加上num_channels->num_classes层，得到分割结果

二、HRNet中应用的结构块

1、普通的3*3的卷积，结构如下：

def conv3x3(in_planes, out_planes, stride=1):
    """3x3 convolution with padding"""
    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
                     padding=1, bias=False)

2、BasicBlock，结构如下：

class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(BasicBlock, self).__init__()
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = nn.BatchNorm2d(planes, momentum=BN_MOMENTUM)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = nn.BatchNorm2d(planes, momentum=BN_MOMENTUM)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)

        return out

3、三层的残差块

expansion的参数，这个参数用来控制卷积的输入输出通道数。

class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes, momentum=BN_MOMENTUM)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
                               padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes, momentum=BN_MOMENTUM)
        self.conv3 = nn.Conv2d(planes, planes * self.expansion, kernel_size=1,
                               bias=False)
        self.bn3 = nn.BatchNorm2d(planes * self.expansion,
                               momentum=BN_MOMENTUM)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)

        return out

三、具体模块

1、HRNet 核心模块类：高分辨率模块（class HighResolutionModule）

实现下图红框中的，branch 并行 多 scale 特征提取 和 末端将 多 scale 特征通过 upsample/downsample 方式融合

class HighResolutionModule(nn.Module):
    def __init__(self, num_branches, blocks, num_blocks, num_inchannels,
                 num_channels, fuse_method, multi_scale_output=True):
        super(HighResolutionModule, self).__init__()
        self._check_branches(
            num_branches, blocks, num_blocks, num_inchannels, num_channels)

        self.num_inchannels = num_inchannels
        self.fuse_method = fuse_method
        self.num_branches = num_branches

        self.multi_scale_output = multi_scale_output

        self.branches = self._make_branches(
            num_branches, blocks, num_blocks, num_channels)
        self.fuse_layers = self._make_fuse_layers()
        self.relu = nn.ReLU(False)

2、check_branches()函数

这个函数的作用是检查，在高分辨率模块中num_branches(int类型)，和len(num_inchannels(里面的元素是int))，和len(num_channels(里面的元素是int))它们三个的值是否相等。

def _check_branches(self, num_branches, blocks, num_blocks,
                        num_inchannels, num_channels):
        if num_branches != len(num_blocks):
            error_msg = 'NUM_BRANCHES({}) <> NUM_BLOCKS({})'.format(
                num_branches, len(num_blocks))
            logger.error(error_msg)
            raise ValueError(error_msg)

        if num_branches != len(num_channels):
            error_msg = 'NUM_BRANCHES({}) <> NUM_CHANNELS({})'.format(
                num_branches, len(num_channels))
            logger.error(error_msg)
            raise ValueError(error_msg)

        if num_branches != len(num_inchannels):
            error_msg = 'NUM_BRANCHES({}) <> NUM_INCHANNELS({})'.format(
                num_branches, len(num_inchannels))
            logger.error(error_msg)
            raise ValueError(error_msg)

3、make_one_branch函数

它的作用就是创建一个新的分支，如图

def _make_one_branch(self, branch_index, block, num_blocks, num_channels,
                         stride=1):
        downsample = None
        if stride != 1 or \
           self.num_inchannels[branch_index] != num_channels[branch_index] * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.num_inchannels[branch_index],
                          num_channels[branch_index] * block.expansion,
                          kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(num_channels[branch_index] * block.expansion,
                            momentum=BN_MOMENTUM),
            )

        layers = []
        layers.append(block(self.num_inchannels[branch_index],
                            num_channels[branch_index], stride, downsample))
        self.num_inchannels[branch_index] = \
            num_channels[branch_index] * block.expansion
        for i in range(1, num_blocks[branch_index]):
            layers.append(block(self.num_inchannels[branch_index],
                                num_channels[branch_index]))

        return nn.Sequential(*layers)

make_branches函数是看看每个stage里面有多少branch，然后有几个就调用几次_make_one_branch函数。

4、forward

def forward(self, x):
        if self.num_branches == 1:
            return [self.branches[0](x[0])]

        for i in range(self.num_branches):
            x[i] = self.branches[i](x[i])

        x_fuse = []
        for i in range(len(self.fuse_layers)):
            y = x[0] if i == 0 else self.fuse_layers[i][0](x[0])
            for j in range(1, self.num_branches):
                if i == j:
                    y = y + x[j]
                else:
                    y = y + self.fuse_layers[i][j](x[j])
            x_fuse.append(self.relu(y))

        return x_fuse

5、构建 multi-scale 特征融合层：fuse_layer函数

代码中，双重循环里面的i代表的当前融合的branch，上面的图我画出了当i=0时，所有的featuremaps都融合到0这个分支的featuremaps上面去，j代表组成融合的featuremaps所对应的branchindex

def _make_fuse_layers(self):
        if self.num_branches == 1:
            return None

        num_branches = self.num_branches
        num_inchannels = self.num_inchannels
        fuse_layers = []
        for i in range(num_branches if self.multi_scale_output else 1):
            fuse_layer = []
            for j in range(num_branches):
                if j > i:
                    fuse_layer.append(nn.Sequential(
                        nn.Conv2d(num_inchannels[j],
                                  num_inchannels[i],
                                  1,
                                  1,
                                  0,
                                  bias=False),
                        nn.BatchNorm2d(num_inchannels[i], 
                                       momentum=BN_MOMENTUM),
                        nn.Upsample(scale_factor=2**(j-i), mode='nearest')))
                elif j == i:
                    fuse_layer.append(None)
                else:
                    conv3x3s = []
                    for k in range(i-j):
                        if k == i - j - 1:
                            num_outchannels_conv3x3 = num_inchannels[i]
                            conv3x3s.append(nn.Sequential(
                                nn.Conv2d(num_inchannels[j],
                                          num_outchannels_conv3x3,
                                          3, 2, 1, bias=False),
                                nn.BatchNorm2d(num_outchannels_conv3x3, 
                                            momentum=BN_MOMENTUM)))
                        else:
                            num_outchannels_conv3x3 = num_inchannels[j]
                            conv3x3s.append(nn.Sequential(
                                nn.Conv2d(num_inchannels[j],
                                          num_outchannels_conv3x3,
                                          3, 2, 1, bias=False),
                                nn.BatchNorm2d(num_outchannels_conv3x3,
                                            momentum=BN_MOMENTUM),
                                nn.ReLU(False)))
                    fuse_layer.append(nn.Sequential(*conv3x3s))
            fuse_layers.append(nn.ModuleList(fuse_layer))

        return nn.ModuleList(fuse_layers)

6、transition_layers函数（上图中画叉的那一个分支）

transition layer 完成 stage 之间连接需要的 两种转换
（1）input channels 转换
（2）feature size downsample

def _make_transition_layer(
            self, num_channels_pre_layer, num_channels_cur_layer):
        num_branches_cur = len(num_channels_cur_layer)
        num_branches_pre = len(num_channels_pre_layer)

        transition_layers = []
        for i in range(num_branches_cur):
            if i < num_branches_pre:
                if num_channels_cur_layer[i] != num_channels_pre_layer[i]:
                    transition_layers.append(nn.Sequential(
                        nn.Conv2d(num_channels_pre_layer[i],
                                  num_channels_cur_layer[i],
                                  3,
                                  1,
                                  1,
                                  bias=False),
                        nn.BatchNorm2d(
                            num_channels_cur_layer[i], momentum=BN_MOMENTUM),
                        nn.ReLU(inplace=True)))
                else:
                    transition_layers.append(None)
            else:
                conv3x3s = []
                for j in range(i+1-num_branches_pre):
                    inchannels = num_channels_pre_layer[-1]
                    outchannels = num_channels_cur_layer[i] \
                        if j == i-num_branches_pre else inchannels
                    conv3x3s.append(nn.Sequential(
                        nn.Conv2d(
                            inchannels, outchannels, 3, 2, 1, bias=False),
                        nn.BatchNorm2d(outchannels, momentum=BN_MOMENTUM),
                        nn.ReLU(inplace=True)))
                transition_layers.append(nn.Sequential(*conv3x3s))

        return nn.ModuleList(transition_layers)

7、构建 stage1 的 layer _make_layer()

stage1 产生 1/4 feature map，没有 branch 分支结构，采用与 resnet 完成一样的 _make_layer() 函数构建层

8、构建 stage 2/3/4 的 layer _make_stage

stage 2/3/4 为 HRNet 核心结构，用到了核心类 HighResolutionModule，内含 make_branches 构建和特征 _make_fuse_layers 模块

四、数据流图