FCOS代码（一） (demo过程)骨干网络结构详解，mask-rcnn ResNet+fpn

代码来自tianzhi0549 / FCOS，

下面是源代码中打印的整体网络结构，然后将其分为ResNet和FPN两部分描述进行描述，config使用的是fcos_imprv_R_50_FPN_1x.yaml，BACKBONE.CONV_BODY=”R-50-FPN-RETINANET”

Sequential(
  (body): ResNet(
    (stem): StemWithFixedBatchNorm(
      (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
      (bn1): FrozenBatchNorm2d()
    )
    (layer1): Sequential(
      (0): BottleneckWithFixedBatchNorm(
        (downsample): Sequential(
          (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (1): FrozenBatchNorm2d()
        )
        (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (1): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (2): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
    )
    (layer2): Sequential(
      (0): BottleneckWithFixedBatchNorm(
        (downsample): Sequential(
          (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
          (1): FrozenBatchNorm2d()
        )
        (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (1): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (2): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (3): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
    )
    (layer3): Sequential(
      (0): BottleneckWithFixedBatchNorm(
        (downsample): Sequential(
          (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
          (1): FrozenBatchNorm2d()
        )
        (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (1): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (2): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (3): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (4): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (5): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
    )
    (layer4): Sequential(
      (0): BottleneckWithFixedBatchNorm(
        (downsample): Sequential(
          (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
          (1): FrozenBatchNorm2d()
        )
        (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (1): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
      (2): BottleneckWithFixedBatchNorm(
        (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): FrozenBatchNorm2d()
        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (bn2): FrozenBatchNorm2d()
        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): FrozenBatchNorm2d()
      )
    )
  )
  (fpn): FPN(
    (fpn_inner2): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1))
    (fpn_layer2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (fpn_inner3): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
    (fpn_layer3): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (fpn_inner4): Conv2d(2048, 256, kernel_size=(1, 1), stride=(1, 1))
    (fpn_layer4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (top_blocks): LastLevelP6P7(
      (p6): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
      (p7): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
    )
  )

1）ResNet部分

上面的ResNet骨干网络采用的是表1中的50-layer结构，对照着ResNet的网络结构表格（见下面表格），接下来细节描述一下：

表1 ResNet结构表

i. StemWithFixedBatchNorm 类：

为表1中conv1 表示的模块外下面的3×3 max pool，由于这个模块各种结构都是通用的，并且接受图片输入（输入通道都是3，输出通道因不同的结构而异），所以名字带有stem（茎干），其代码定义如下，从中可以看出与表1conv1 的部分完全吻合(forward conv1 后接F.max_pool2d)

class BaseStem(nn.Module):
    def __init__(self, cfg, norm_func):
        super(BaseStem, self).__init__()

        out_channels = cfg.MODEL.RESNETS.STEM_OUT_CHANNELS

        self.conv1 = Conv2d(
            3, out_channels, kernel_size=7, stride=2, padding=3, bias=False
        )
        self.bn1 = norm_func(out_channels)

        for l in [self.conv1,]:
            nn.init.kaiming_uniform_(l.weight, a=1)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = F.relu_(x)
        x = F.max_pool2d(x, kernel_size=3, stride=2, padding=1)
        return x
class StemWithFixedBatchNorm(BaseStem):
    def __init__(self, cfg):
        super(StemWithFixedBatchNorm, self).__init__(
            cfg, norm_func=FrozenBatchNorm2d
        )

ii. FrozenBatchNorm2d 类

定义如下，一些相关参数的形状如后面注释所示（n值为前面相邻卷积的输出通道数），与正常的BatchNorm 层的作用一样，使一批Batch 的 feature map 满足均值为0，方差为1的分布规律。区别是它使用了 self.register_buffer ，参数固定，训练时不更新，参考pytorch 中register_buffer（）

class FrozenBatchNorm2d(nn.Module):
"""
    BatchNorm2d where the batch statistics and the affine parameters
    are fixed
"""

    def __init__(self, n):  # n=64
        super(FrozenBatchNorm2d, self).__init__()
        self.register_buffer("weight", torch.ones(n))
        self.register_buffer("bias", torch.zeros(n))
        self.register_buffer("running_mean", torch.zeros(n))
        self.register_buffer("running_var", torch.ones(n))

    def forward(self, x):  # x= {Tensor:(1,64,400,560)}
        scale = self.weight * self.running_var.rsqrt()  # Tensor:(64,)
        bias = self.bias - self.running_mean * scale  # Tensor:(64,)
        scale = scale.reshape(1, -1, 1, 1)  # Tensor: (1,64,1,1)
        bias = bias.reshape(1, -1, 1, 1)  # Tensor: (1,64,1,1)
        return x * scale + bias

iii. BottleneckWithFixedBatchNorm 类

这个就是定义构成ResNet的基本模块类（如下面代码所示），从文章开头所示的整体网络结构中可以看见，每层（不同的layer表示）开头都有downsample 结构，这是由于ResNet残差网络结构由于需要每层的输入和输出进行加和，如果该层的输入通道数与输出通道数不一样（即形状不一致），则需要对输入进行一个转变使得通道数保持与输出通道数一致。可以参考CV脱坑指南（二）：ResNet·downsample详解

class Bottleneck(nn.Module):
    def __init__(
        self,
        in_channels,
        bottleneck_channels,
        out_channels,
        num_groups,
        stride_in_1x1,
        stride,
        dilation,
        norm_func,
        dcn_config
    ):
        super(Bottleneck, self).__init__()

        self.downsample = None  # 输入通道与输出通道不相等则采用这个
        if in_channels != out_channels:
            down_stride = stride if dilation == 1 else 1
            self.downsample = nn.Sequential(
                Conv2d(
                    in_channels, out_channels,
                    kernel_size=1, stride=down_stride, bias=False
                ),
                norm_func(out_channels),
            )
            for modules in [self.downsample,]:
                for l in modules.modules():
                    if isinstance(l, Conv2d):
                        nn.init.kaiming_uniform_(l.weight, a=1)

        if dilation > 1:
            stride = 1 # reset to be 1

        # The original MSRA ResNet models have stride in the first 1x1 conv
        # The subsequent fb.torch.resnet and Caffe2 ResNe[X]t implementations have
        # stride in the 3x3 conv
        stride_1x1, stride_3x3 = (stride, 1) if stride_in_1x1 else (1, stride)

        self.conv1 = Conv2d(
            in_channels,
            bottleneck_channels,
            kernel_size=1,
            stride=stride_1x1,
            bias=False,
        )
        self.bn1 = norm_func(bottleneck_channels)
        # TODO: specify init for the above
        with_dcn = dcn_config.get("stage_with_dcn", False)
        if with_dcn:
            deformable_groups = dcn_config.get("deformable_groups", 1)
            with_modulated_dcn = dcn_config.get("with_modulated_dcn", False)
            self.conv2 = DFConv2d(
                bottleneck_channels,
                bottleneck_channels,
                with_modulated_dcn=with_modulated_dcn,
                kernel_size=3,
                stride=stride_3x3,
                groups=num_groups,
                dilation=dilation,
                deformable_groups=deformable_groups,
                bias=False
            )
        else:
            self.conv2 = Conv2d(
                bottleneck_channels,
                bottleneck_channels,
                kernel_size=3,
                stride=stride_3x3,
                padding=dilation,
                bias=False,
                groups=num_groups,
                dilation=dilation
            )
            nn.init.kaiming_uniform_(self.conv2.weight, a=1)

        self.bn2 = norm_func(bottleneck_channels)

        self.conv3 = Conv2d(
            bottleneck_channels, out_channels, kernel_size=1, bias=False
        )
        self.bn3 = norm_func(out_channels)

        for l in [self.conv1, self.conv3,]:
            nn.init.kaiming_uniform_(l.weight, a=1)

    def forward(self, x):
        identity = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = F.relu_(out)

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

        out0 = self.conv3(out)
        out = self.bn3(out0)

        if self.downsample is not None:  # 采用downsample使输入和输出通道数相等
            identity = self.downsample(x)

        out += identity  # 残差结构 输入与输出 相加
        out = F.relu_(out)

        return out
class BottleneckWithFixedBatchNorm(Bottleneck):
    def __init__(
        self,
        in_channels,
        bottleneck_channels,
        out_channels,
        num_groups=1,
        stride_in_1x1=True,
        stride=1,
        dilation=1,
        dcn_config=None
    ):
        super(BottleneckWithFixedBatchNorm, self).__init__(
            in_channels=in_channels,
            bottleneck_channels=bottleneck_channels,
            out_channels=out_channels,
            num_groups=num_groups,
            stride_in_1x1=stride_in_1x1,
            stride=stride,
            dilation=dilation,
            norm_func=FrozenBatchNorm2d,
            dcn_config=dcn_config
        )

iv. ResNet 的定义

至此，ResNet的定义如下，每一层的输出都保存在了outputs的列表里（看注释），一共4个layers（见开头打印的部分）所以共4个输出，并且每个输出的size依次缩小一倍（2倍下采样）。它的输出接下来会与FPN网络相结合。

class ResNet(nn.Module):
    def __init__(self, cfg):
        super(ResNet, self).__init__()

        # If we want to use the cfg in forward(), then we should make a copy
        # of it and store it for later use:
        # self.cfg = cfg.clone()

        # Translate string names to implementations
        stem_module = _STEM_MODULES[cfg.MODEL.RESNETS.STEM_FUNC]
        stage_specs = _STAGE_SPECS[cfg.MODEL.BACKBONE.CONV_BODY]
        transformation_module = _TRANSFORMATION_MODULES[cfg.MODEL.RESNETS.TRANS_FUNC]

        # Construct the stem module
        self.stem = stem_module(cfg)

        # Constuct the specified ResNet stages
        num_groups = cfg.MODEL.RESNETS.NUM_GROUPS  # 1
        width_per_group = cfg.MODEL.RESNETS.WIDTH_PER_GROUP  # 64
        in_channels = cfg.MODEL.RESNETS.STEM_OUT_CHANNELS  # 2048
        stage2_bottleneck_channels = num_groups * width_per_group  # 64
        stage2_out_channels = cfg.MODEL.RESNETS.RES2_OUT_CHANNELS  # 256
        self.stages = []  # ['layer1','layer2','layer3','layer4']
        self.return_features = {}  # {'layer1':True,'layer2':True,'layer3':True,'layer4':True}
        for stage_spec in stage_specs:
            name = "layer" + str(stage_spec.index)
            stage2_relative_factor = 2 ** (stage_spec.index - 1)
            bottleneck_channels = stage2_bottleneck_channels * stage2_relative_factor
            out_channels = stage2_out_channels * stage2_relative_factor
            stage_with_dcn = cfg.MODEL.RESNETS.STAGE_WITH_DCN[stage_spec.index - 1]
            module = _make_stage(
                transformation_module,
                in_channels,
                bottleneck_channels,
                out_channels,
                stage_spec.block_count,
                num_groups,
                cfg.MODEL.RESNETS.STRIDE_IN_1X1,
                first_stride=int(stage_spec.index > 1) + 1,
                dcn_config={
                    "stage_with_dcn": stage_with_dcn,
                    "with_modulated_dcn": cfg.MODEL.RESNETS.WITH_MODULATED_DCN,
                    "deformable_groups": cfg.MODEL.RESNETS.DEFORMABLE_GROUPS,
                }
            )
            in_channels = out_channels
            self.add_module(name, module)
            self.stages.append(name)
            self.return_features[name] = stage_spec.return_features

        # Optionally freeze (requires_grad=False) parts of the backbone
        self._freeze_backbone(cfg.MODEL.BACKBONE.FREEZE_CONV_BODY_AT)

    def _freeze_backbone(self, freeze_at):
        if freeze_at < 0:
            return
        for stage_index in range(freeze_at):
            if stage_index == 0:
                m = self.stem  # stage 0 is the stem
            else:
                m = getattr(self, "layer" + str(stage_index))
            for p in m.parameters():
                p.requires_grad = False

    def forward(self, x):
        outputs = []  # [Tensor:(1,256,200,280), Tensor:(1,512,100,140), Tensor:(1,1024,70,70), Tensor:(1,2048,25,35)]
        x = self.stem(x)
        for stage_name in self.stages:
            x = getattr(self, stage_name)(x)
            if self.return_features[stage_name]:
                outputs.append(x)
        return outputs

其中，state_specs 值如下图所示，它包含的是定义的StageSpec字典类。namedtuple的用法参考Python的namedtuple使用详解。

StageSpec = namedtuple(
    "StageSpec",
    [
        "index",  # Index of the stage, eg 1, 2, ..,. 5
        "block_count",  # Number of residual blocks in the stage
        "return_features",  # True => return the last feature map from this stage
    ],
)

2） FPN部分

FPN网络的整体构建如下，关于FPN网络的理解参考FPN详解

in_channels_stage2 = cfg.MODEL.RESNETS.RES2_OUT_CHANNELS  # 256
out_channels = cfg.MODEL.RESNETS.BACKBONE_OUT_CHANNELS  # 256
in_channels_p6p7 = in_channels_stage2 * 8 if cfg.MODEL.RETINANET.USE_C5 \
    else out_channels  # 256
fpn = fpn_module.FPN(
    in_channels_list=[
        0,
        in_channels_stage2 * 2,
        in_channels_stage2 * 4,
        in_channels_stage2 * 8,
    ],
    out_channels=out_channels,
    conv_block=conv_with_kaiming_uniform(
        cfg.MODEL.FPN.USE_GN, cfg.MODEL.FPN.USE_RELU
    ),
    top_blocks=fpn_module.LastLevelP6P7(in_channels_p6p7, out_channels),
)
model = nn.Sequential(OrderedDict([("body", body), ("fpn", fpn)]))
model.out_channels = out_channels
return model

下面细节分析以下这个构建过程的要素

i. conv_block — 传入的参数

可以看到conv_block = conv_with_kaiming_uniform（cfg.MODEL.FPN.USE_GN, cfg.MODEL.FPN.USE_RELU），其中cfg.MODEL.FPN.USE_GN=False，cfg.MODEL.FPN.USE_RELU=False，conv_with_kaiming_uniform函数如下所示，它的返回值是一个函数 make_conv, 是组成FPN网络的基本模块。其中nn.init.kaiming_uniform_ 是一种权重参数的初始化方法，名为kaiming(由何凯明提出)，参考神经网络权重初始化代码 init.kaiming_uniform_和kaiming_normal_。 use_gn中的gn指的是group norm 归一化操作，将channel方向分group，然后每个group内做归一化，算(C//G) _H_W的均值，这样与batchsize无关，不受其约束，参考PyTorch学习之归一化层（BatchNorm、LayerNorm、InstanceNorm、GroupNorm）

def conv_with_kaiming_uniform(use_gn=False, use_relu=False):
    def make_conv(
        in_channels, out_channels, kernel_size, stride=1, dilation=1
    ):
        conv = Conv2d(
            in_channels,
            out_channels,
            kernel_size=kernel_size,
            stride=stride,
            padding=dilation * (kernel_size - 1) // 2,
            dilation=dilation,
            bias=False if use_gn else True
        )
        # Caffe2 implementation uses XavierFill, which in fact
        # corresponds to kaiming_uniform_ in PyTorch
        nn.init.kaiming_uniform_(conv.weight, a=1)
        if not use_gn:
            nn.init.constant_(conv.bias, 0)
        module = [conv,]
        if use_gn:
            module.append(group_norm(out_channels))
        if use_relu:
            module.append(nn.ReLU(inplace=True))
        if len(module) > 1:
            return nn.Sequential(*module)
        return conv

    return make_conv

ii. top_blocks –参数

其有两种定义，一种是fpn_module.LastLevelMaxPool()

定义如下，就是执行max_pool2d操作，只不过使用的是torch.nn.function里的，参数依次为 input，kernel_size，stride，padding，详见pytorch官网，不过它的作用与torch.nn.MaxPool2d一样，对于输入信号的输入通道，提供2维最大池化（max pooling）操作。

class LastLevelMaxPool(nn.Module):
    def forward(self, x):
        return [F.max_pool2d(x, 1, 2, 0)]

另一种（也是这次使用的）是LastLevelP6P7，定义如下。表示最后一层不需要池化，而是在FPN上另外加了P6、P7两层。

class LastLevelP6P7(nn.Module):
"""
    This module is used in RetinaNet to generate extra layers, P6 and P7.

"""
    def __init__(self, in_channels, out_channels):
        super(LastLevelP6P7, self).__init__()
        self.p6 = nn.Conv2d(in_channels, out_channels, 3, 2, 1)
        self.p7 = nn.Conv2d(out_channels, out_channels, 3, 2, 1)
        for module in [self.p6, self.p7]:
            nn.init.kaiming_uniform_(module.weight, a=1)
            nn.init.constant_(module.bias, 0)
        self.use_P5 = in_channels == out_channels  # True

    def forward(self, c5, p5):
        x = p5 if self.use_P5 else c5
        p6 = self.p6(x)
        p7 = self.p7(F.relu(p6))
        return [p6, p7]

iii. FPN类

至此，最终的FPN类定义如下，注意results（最终输出5个Tesnsor，看注释）里的结果，按次序size依次缩小2倍，而且注意last_inner = inner_lateral + inner_top_down 这步操作，就是FPN的加和操作。其中的inner_lateral是对来自FPN网络左边的top-down结构（按次序来的，从上到下）采用1×1的卷积进行降维处理，inner_top_down是对FPN网络右边的结构进行上采样操作（利用F.interpolate实现，参考F.interpolate——数组采样操作），然后将两者相加（对应元素相加），最后进行3×3的卷积操作（参考FPN理解）。这里是从上到下进行的（从顶部先输出小尺寸的特征图），只不过利用了列表的insert函数（参考python insert()函数解析）依次把后一个结果插到之前位置。last_results = self.top_blocks(x[-1], results[-1]) 这里就是采用外加的P6P7层，x[-1], results[-1]是传入的c5（ResNet提取特征的最后一层）和p5（FPN在加入P6和P7之前的次序最后的输出，size最小的）参数，最终选取的是p5。

class FPN(nn.Module):
"""
    Module that adds FPN on top of a list of feature maps.

    The feature maps are currently supposed to be in increasing depth
    order, and must be consecutive
"""

    def __init__(
        self, in_channels_list, out_channels, conv_block, top_blocks=None
    ):
"""
        Arguments:
            in_channels_list (list[int]): number of channels for each feature map that
                will be fed
            out_channels (int): number of channels of the FPN representation
            top_blocks (nn.Module or None): if provided, an extra operation will
                be performed on the output of the last (smallest resolution)
                FPN output, and the result will extend the result list
"""
        super(FPN, self).__init__()
        self.inner_blocks = []  # ['fpn_inner2','fpn_inner3','fpn_inner4']
        self.layer_blocks = []  # ['fpn_layer2','fpn_layer2','fpn_layer2']
        for idx, in_channels in enumerate(in_channels_list, 1):
            inner_block = "fpn_inner{}".format(idx)
            layer_block = "fpn_layer{}".format(idx)

            if in_channels == 0:
                continue
            inner_block_module = conv_block(in_channels, out_channels, 1)
            layer_block_module = conv_block(out_channels, out_channels, 3, 1)
            self.add_module(inner_block, inner_block_module)
            self.add_module(layer_block, layer_block_module)
            self.inner_blocks.append(inner_block)
            self.layer_blocks.append(layer_block)
        self.top_blocks = top_blocks  # LastLevelP6P7

    def forward(self, x):
"""
        Arguments:
            x (list[Tensor]): feature maps for each feature level.

        Returns:
            results (tuple[Tensor]): feature maps after FPN layers.

                They are ordered from highest resolution first.

"""
        last_inner = getattr(self, self.inner_blocks[-1])(x[-1])
        results = []  # [Tensor:(1,256,100,140),Tensor:(1,256,50,70),Tensor:(1,256,25,35),Tensor:(1,256,13,18),Tensor:(1,256,7,9)]
        results.append(getattr(self, self.layer_blocks[-1])(last_inner))
        for feature, inner_block, layer_block in zip(
            x[:-1][::-1], self.inner_blocks[:-1][::-1], self.layer_blocks[:-1][::-1]
        ):
            if not inner_block:
                continue
            # inner_top_down = F.interpolate(last_inner, scale_factor=2, mode="nearest")
            inner_lateral = getattr(self, inner_block)(feature)
            inner_top_down = F.interpolate(
                last_inner, size=(int(inner_lateral.shape[-2]), int(inner_lateral.shape[-1])),
                mode='nearest'
            )
            last_inner = inner_lateral + inner_top_down
            results.insert(0, getattr(self, layer_block)(last_inner))

        if isinstance(self.top_blocks, LastLevelP6P7):  # before this list of results is 3, then add 2 become 5
            last_results = self.top_blocks(x[-1], results[-1])
            results.extend(last_results)
        elif isinstance(self.top_blocks, LastLevelMaxPool):
            last_results = self.top_blocks(results[-1])
            results.extend(last_results)

        return tuple(results)

Original: https://blog.csdn.net/allrubots/article/details/124842487
Author: 匿名的魔术师
Title: FCOS代码（一） (demo过程)骨干网络结构详解，mask-rcnn ResNet+fpn