CenterNet网络模型代码_3-pose_dla_dcn.py

这里主要针对在CenterNet中表现还可以的改进过的 dla 网络进行分析，主要还是结合代码进行说明。下面就是代码：

# BN 动量参数
BN_MOMENTUM = 0.1
# 获取打印日志对象
logger = logging.getLogger(__name__)
# 获取预训练网络
def get_model_url(data='imagenet', name='dla34', hash='ba72cf86'):
    return join('http://dl.yf.io/dla/models', data, '{}-{}.pth'.format(name, hash))
# 3*3 卷积
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)

#基础模块（Res18 34中运用）
#3*3 bn relu -> 3*3 bn relu + x(res) 
class BasicBlock(nn.Module):
    def __init__(self, inplanes, planes, stride=1, dilation=1):
        super(BasicBlock, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=3,
                               stride=stride, padding=dilation,
                               bias=False, dilation=dilation)
        self.bn1 = nn.BatchNorm2d(planes, momentum=BN_MOMENTUM)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3,
                               stride=1, padding=dilation,
                               bias=False, dilation=dilation)
        self.bn2 = nn.BatchNorm2d(planes, momentum=BN_MOMENTUM)
        self.stride = stride

    def forward(self, x, residual=None):
        if residual is None:
            residual = x

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

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

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

        return out
# 残差瓶颈层(50以上的层次用)
#Resnet bottleneck 
# 这个瓶颈层其实和 ResNet中的瓶颈层类似，只不过其宽度不同，这里更宽
class Bottleneck(nn.Module):
    expansion = 2#这里的膨胀率是2，也就是让网络更宽
    # 初始化
    def __init__(self, inplanes, planes, stride=1, dilation=1):
        super(Bottleneck, self).__init__()
        expansion = Bottleneck.expansion
        bottle_planes = planes // expansion
        self.conv1 = nn.Conv2d(inplanes, bottle_planes,
                               kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(bottle_planes, momentum=BN_MOMENTUM)
        self.conv2 = nn.Conv2d(bottle_planes, bottle_planes, kernel_size=3,
                               stride=stride, padding=dilation,
                               bias=False, dilation=dilation)
        self.bn2 = nn.BatchNorm2d(bottle_planes, momentum=BN_MOMENTUM)
        self.conv3 = nn.Conv2d(bottle_planes, planes,
                               kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes, momentum=BN_MOMENTUM)
        self.relu = nn.ReLU(inplace=True)
        self.stride = stride

    def forward(self, x, residual=None):
        if residual is None:
            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)

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

        return out

#ResneXt bottleneck 
class BottleneckX(nn.Module):
    expansion = 2
    cardinality = 32

    def __init__(self, inplanes, planes, stride=1, dilation=1):
        super(BottleneckX, self).__init__()
        cardinality = BottleneckX.cardinality
        # dim = int(math.floor(planes * (BottleneckV5.expansion / 64.0)))
        # bottle_planes = dim * cardinality
        bottle_planes = planes * cardinality // 32
        self.conv1 = nn.Conv2d(inplanes, bottle_planes,kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(bottle_planes, momentum=BN_MOMENTUM)
        # 使用了组卷积，划分方式为 32 组，如果输入的通道是128，输出通道是256 则卷积核会变成 32*8*3*3*4，即每次只考虑在输入4 个通道上 做卷积，输出为8 通道，然后再把 32个输出的 8 通道特征融合(concat)为 256维
        self.conv2 = nn.Conv2d(bottle_planes, bottle_planes, kernel_size=3,stride=stride, padding=dilation, bias=False,dilation=dilation, groups=cardinality)
        self.bn2 = nn.BatchNorm2d(bottle_planes, momentum=BN_MOMENTUM)
        self.conv3 = nn.Conv2d(bottle_planes, planes,
                               kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes, momentum=BN_MOMENTUM)
        self.relu = nn.ReLU(inplace=True)
        self.stride = stride

    def forward(self, x, residual=None):
        if residual is None:
            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)

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

        return out

# 根 1*1 (in*out) 调整维度的卷积操作，可以使用残差模块
class Root(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size, residual):
        super(Root, self).__init__()
        self.conv = nn.Conv2d(
            in_channels, out_channels, 1,
            stride=1, bias=False, padding=(kernel_size - 1) // 2)
        self.bn = nn.BatchNorm2d(out_channels, momentum=BN_MOMENTUM)
        self.relu = nn.ReLU(inplace=True)
        self.residual = residual

    def forward(self, *x):
        children = x
        x = self.conv(torch.cat(x, 1))
        x = self.bn(x)
        if self.residual:
            x += children[0]
        x = self.relu(x)

        return x

# 3*3 conv stack
class Tree(nn.Module):
    def __init__(self, levels, block, in_channels, out_channels, stride=1,
                 level_root=False, root_dim=0, root_kernel_size=1,
                 dilation=1, root_residual=False):
        super(Tree, self).__init__()
        if root_dim == 0:
            root_dim = 2 * out_channels
        if level_root:
            root_dim += in_channels
        if levels == 1:
            self.tree1 = block(in_channels, out_channels, stride,
                               dilation=dilation)
            self.tree2 = block(out_channels, out_channels, 1,
                               dilation=dilation)
        else:
        # 堆叠的 BasicBlock 树结构
            self.tree1 = Tree(levels - 1, block, in_channels, out_channels,stride, root_dim=0,root_kernel_size=root_kernel_size,
                              dilation=dilation, root_residual=root_residual)
            self.tree2 = Tree(levels - 1, block, out_channels, out_channels,
                              root_dim=root_dim + out_channels,
                              root_kernel_size=root_kernel_size,
                              dilation=dilation, root_residual=root_residual)
        if levels == 1:
            self.root = Root(root_dim, out_channels, root_kernel_size,
                             root_residual)
        self.level_root = level_root
        self.root_dim = root_dim
        self.downsample = None
        self.project = None
        self.levels = levels
        if stride > 1:
            self.downsample = nn.MaxPool2d(stride, stride=stride)
        if in_channels != out_channels:
            self.project = nn.Sequential(
                nn.Conv2d(in_channels, out_channels,
                          kernel_size=1, stride=1, bias=False),
                nn.BatchNorm2d(out_channels, momentum=BN_MOMENTUM)
            )

    def forward(self, x, residual=None, children=None):
        children = [] if children is None else children
        #下采样为True则进行下采样，不然就是 x 本身
        bottom = self.downsample(x) if self.downsample else x
        # 是否使用1*1卷积进行维度改变(Cin!=Cout)
        residual = self.project(bottom) if self.project else bottom
        if self.level_root:
            children.append(bottom)
        x1 = self.tree1(x, residual)# 通过一个残差块
        if self.levels == 1:
            x2 = self.tree2(x1)#通过第二个残差块
            #进行组合x2和x1,进行1*1卷积
            x = self.root(x2, x1, *children)
        else:
            children.append(x1)
            x = self.tree2(x1, children=children)
        return x

#DLA network backbone
class DLA(nn.Module):
    def __init__(self, levels, channels, num_classes=1000,
                 block=BasicBlock, residual_root=False, linear_root=False):
        super(DLA, self).__init__()
        self.channels = channels#[16,32,64,128,256,512]
        self.num_classes = num_classes#94
        # pre-precess 7*7 conv
        self.base_layer = nn.Sequential(
            nn.Conv2d(3, channels[0], kernel_size=7, stride=1,
                      padding=3, bias=False),
            nn.BatchNorm2d(channels[0], momentum=BN_MOMENTUM),
            nn.ReLU(inplace=True))
        self.level0 = self._make_conv_level(
            channels[0], channels[0], levels[0])
        self.level1 = self._make_conv_level(channels[0], channels[1], levels[1], stride=2)
        self.level2 = Tree(levels[2], block, channels[1], channels[2], 2,level_root=False,root_residual=residual_root)
        self.level3 = Tree(levels[3], block, channels[2], channels[3], 2,level_root=True, root_residual=residual_root)
        self.level4 = Tree(levels[4], block, channels[3], channels[4], 2,level_root=True, root_residual=residual_root)
        self.level5 = Tree(levels[5], block, channels[4], channels[5], 2,level_root=True, root_residual=residual_root)

        # for m in self.modules():
        #     if isinstance(m, nn.Conv2d):
        #         n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
        #         m.weight.data.normal_(0, math.sqrt(2. / n))
        #     elif isinstance(m, nn.BatchNorm2d):
        #         m.weight.data.fill_(1)
        #         m.bias.data.zero_()
   # 构建每阶段的卷积
    def _make_level(self, block, inplanes, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or inplanes != planes:
            downsample = nn.Sequential(
                nn.MaxPool2d(stride, stride=stride),
                nn.Conv2d(inplanes, planes,
                          kernel_size=1, stride=1, bias=False),
                nn.BatchNorm2d(planes, momentum=BN_MOMENTUM),
            )
        layers = []
        layers.append(block(inplanes, planes, stride, downsample=downsample))
        for i in range(1, blocks):
            layers.append(block(inplanes, planes))

        return nn.Sequential(*layers)
    # 3*3 conv 
    def _make_conv_level(self, inplanes, planes, convs, stride=1, dilation=1):
        modules = []
        for i in range(convs):
            modules.extend([
                nn.Conv2d(inplanes, planes, kernel_size=3,
                          stride=stride if i == 0 else 1,
                          padding=dilation, bias=False, dilation=dilation),
                nn.BatchNorm2d(planes, momentum=BN_MOMENTUM),
                nn.ReLU(inplace=True)])
            inplanes = planes
        return nn.Sequential(*modules)

    def forward(self, x):
        y = []
        x = self.base_layer(x)
        for i in range(6):
            x = getattr(self, 'level{}'.format(i))(x)
            y.append(x)
        return y#levl0-5 out put
    # 加载预训练模型
    def load_pretrained_model(self, data='imagenet', name='dla34', hash='ba72cf86'):
        # fc = self.fc
        if name.endswith('.pth'):
            model_weights = torch.load(data + name)
        else:
            model_url = get_model_url(data, name, hash)
            model_weights = model_zoo.load_url(model_url)
        num_classes = len(model_weights[list(model_weights.keys())[-1]])
        self.fc = nn.Conv2d(
            self.channels[-1], num_classes,
            kernel_size=1, stride=1, padding=0, bias=True)
        self.load_state_dict(model_weights)
        # self.fc = fc

# 定义dla的一些模块配置
def dla34(pretrained=True, **kwargs):  # DLA-34
    model = DLA([1, 1, 1, 2, 2, 1],
                [16, 32, 64, 128, 256, 512],
                block=BasicBlock, **kwargs)
    if pretrained:
        model.load_pretrained_model(data='imagenet', name='dla34', hash='ba72cf86')
    return model
# 返回自身数据
class Identity(nn.Module):

    def __init__(self):
        super(Identity, self).__init__()

    def forward(self, x):
        return x

# 初始化权重
def fill_fc_weights(layers):
    for m in layers.modules():
        if isinstance(m, nn.Conv2d):
            if m.bias is not None:
                nn.init.constant_(m.bias, 0)

# 初始化权重
def fill_up_weights(up):
    w = up.weight.data
    f = math.ceil(w.size(2) / 2)
    c = (2 * f - 1 - f % 2) / (2. * f)
    for i in range(w.size(2)):
        for j in range(w.size(3)):
            w[0, 0, i, j] = \
                (1 - math.fabs(i / f - c)) * (1 - math.fabs(j / f - c))
    for c in range(1, w.size(0)):
        w[c, 0, :, :] = w[0, 0, :, :]

# 可变形卷积
class DeformConv(nn.Module):
    def __init__(self, chi, cho):
        super(DeformConv, self).__init__()
        self.actf = nn.Sequential(
            nn.BatchNorm2d(cho, momentum=BN_MOMENTUM),
            nn.ReLU(inplace=True)
        )
        self.conv = DCN(chi, cho, kernel_size=(3,3), stride=1, padding=1, dilation=1, deformable_groups=1)

    def forward(self, x):
        x = self.conv(x)
        x = self.actf(x)
        return x

#上采样
class IDAUp(nn.Module):

    def __init__(self, o, channels, up_f):
        super(IDAUp, self).__init__()
        for i in range(1, len(channels)):
            c = channels[i]
            f = int(up_f[i])  
            proj = DeformConv(c, o)#利用可变形卷积进行降维
            node = DeformConv(o, o)#利用可变形卷积进行调整特征
            #nn.ConvTranspose2d(in_channels, out_channels, kernel_size, stride=1, padding=0, 
            # output_padding=0, groups=1, bias=True, dilation=1)    
            up = nn.ConvTranspose2d(o, o, f * 2, stride=f, 
                                    padding=f // 2, output_padding=0,
                                    groups=o, bias=False)
            fill_up_weights(up)

            setattr(self, 'proj_' + str(i), proj)
            setattr(self, 'up_' + str(i), up)
            setattr(self, 'node_' + str(i), node)
                 
        
    def forward(self, layers, startp, endp):
        for i in range(startp + 1, endp):
            upsample = getattr(self, 'up_' + str(i - startp))
            project = getattr(self, 'proj_' + str(i - startp))
            layers[i] = upsample(project(layers[i]))
            node = getattr(self, 'node_' + str(i - startp))
            layers[i] = node(layers[i] + layers[i - 1])


#upsampling
class DLAUp(nn.Module):
    def __init__(self, startp, channels, scales, in_channels=None):
        super(DLAUp, self).__init__()
        self.startp = startp# 2
        if in_channels is None:
            in_channels = channels
        self.channels = channels#[64, 128, 256, 512]
        channels = list(channels)#[64, 128, 256, 512]
        scales = np.array(scales, dtype=int)#[1,2,4,8]
        for i in range(len(channels) - 1):#[0,1,2]
            j = -i - 2#[-2,-3,-4] IDAUp:
            setattr(self, 'ida_{}'.format(i),
                    IDAUp(channels[j], in_channels[j:],
                          scales[j:] // scales[j]))#set the ida_* 为IDAUp(channels[j], in_channels[j:],scales[j:] // scales[j])
            scales[j + 1:] = scales[j]
            in_channels[j + 1:] = [channels[j] for _ in channels[j + 1:]]

    def forward(self, layers):
        out = [layers[-1]] # start with 32
        for i in range(len(layers) - self.startp - 1):
            ida = getattr(self, 'ida_{}'.format(i))
            ida(layers, len(layers) -i - 2, len(layers))
            out.insert(0, layers[-1])
        return out

#线性插值
class Interpolate(nn.Module):
    def __init__(self, scale, mode):
        super(Interpolate, self).__init__()
        self.scale = scale
        self.mode = mode
        
    def forward(self, x):
        x = F.interpolate(x, scale_factor=self.scale, mode=self.mode, align_corners=False)
        return x

# 构架全部的 改进 DLA 结构
class DLASeg(nn.Module):
    def __init__(self, base_name, heads, pretrained, down_ratio, final_kernel,
                 last_level, head_conv, out_channel=0):
        super(DLASeg, self).__init__()
        assert down_ratio in [2, 4, 8, 16]# down_ratio=4 default
        self.first_level = int(np.log2(down_ratio)) # first_level=2
        self.last_level = last_level # deafult 5
        self.base = globals()[base_name](pretrained=pretrained)#base_name dla34
        channels = self.base.channels#[16, 32, 64, 128, 256, 512]
        scales = [2 ** i for i in range(len(channels[self.first_level:]))]#[1,2,4,8]
        self.dla_up = DLAUp(self.first_level, channels[self.first_level:], scales)

        if out_channel == 0:
            out_channel = channels[self.first_level]

        self.ida_up = IDAUp(out_channel, channels[self.first_level:self.last_level], 
                            [2 ** i for i in range(self.last_level - self.first_level)])
        
        self.heads = heads#{'hm':94,'reg':2,'wh':2} 
        for head in self.heads:
            classes = self.heads[head]
            if head_conv > 0:#head_conv 256
              #channels[self.first_level]:64
              # 3*3-->Relu-->1*1
              fc = nn.Sequential(
                  nn.Conv2d(channels[self.first_level], head_conv,
                    kernel_size=3, padding=1, bias=True),
                  nn.ReLU(inplace=True),
                  nn.Conv2d(head_conv, classes, 
                    kernel_size=final_kernel, stride=1, 
                    padding=final_kernel // 2, bias=True))
              if 'hm' in head:
                fc[-1].bias.data.fill_(-2.19)
              else:
                fill_fc_weights(fc)
            else:
              fc = nn.Conv2d(channels[self.first_level], classes, 
                  kernel_size=final_kernel, stride=1, 
                  padding=final_kernel // 2, bias=True)
              if 'hm' in head:
                fc.bias.data.fill_(-2.19)
              else:
                fill_fc_weights(fc)
            self.__setattr__(head, fc)

    def forward(self, x):
        x = self.base(x)#dla34 output
        x = self.dla_up(x)# dla upsampling

        y = []
        for i in range(self.last_level - self.first_level):
            y.append(x[i].clone())
        self.ida_up(y, 0, len(y))# [64,128,256] stage2_4 upsample directly

        z = {}
        for head in self.heads:
            z[head] = self.__getattr__(head)(y[-1])
        return [z]
 # 模型主函数
def get_pose_net(num_layers, heads, head_conv=256, down_ratio=4):
  model = DLASeg('dla{}'.format(num_layers), heads,
                 pretrained=True,
                 down_ratio=down_ratio,
                 final_kernel=1,
                 last_level=5,
                 head_conv=head_conv)
  return model

代码可能不容易搞懂，因为里面加了递归，网络结构打印信息为,这个为两个堆叠的块结构：

以上是关于CenterNet中用的改进过后的 DLA34的代码，后来会在对该网络结构进行一个细致文章分析