YOLOv9 网络结构 backbone 中的 RepNCSPELAN4 和 ADown 模块

看了YOLOv9的论文和代码，整理了一下网络中用到的模块，以方便理解和日后参考使用。
yolov9论文：YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information
yolov9源码：https://github.com/WongKinYiu/yolov9

YOLOv9整体网络结构图

这里参考 https://blog.csdn.net/qq_44029998/article/details/136324589

backbone中的 RepNCSPELAN4 模块结构

1、模块的参数输入[c1, c2, c3, c4, c5=1]，c1为模块的输入channel（上一模块的输出）, c2为模块的输出channel

2、RepNCSPELAN4模块的整体结构，结合了CSP模块和ELAN模块

这两个模块对应的两篇参考文献
《CSPNet: A New Backbone that can Enhance Learning Capability of CNN》
《Designing Network Design Strategies Through Gradient Path Analysis》

3、RepNCSPELAN4 模块代码

class RepNCSPELAN4(nn.Module):
    # csp-elan
    def __init__(self, c1, c2, c3, c4, c5=1):  # ch_in, ch_out, number, shortcut, groups, expansion
        super().__init__()
        self.c = c3//2
        self.cv1 = Conv(c1, c3, 1, 1)
        self.cv2 = nn.Sequential(RepNCSP(c3//2, c4, c5), Conv(c4, c4, 3, 1))
        self.cv3 = nn.Sequential(RepNCSP(c4, c4, c5), Conv(c4, c4, 3, 1))
        self.cv4 = Conv(c3+(2*c4), c2, 1, 1)

    def forward(self, x):
        y = list(self.cv1(x).chunk(2, 1))
        y.extend((m(y[-1])) for m in [self.cv2, self.cv3])
        return self.cv4(torch.cat(y, 1))

    def forward_split(self, x):
        y = list(self.cv1(x).split((self.c, self.c), 1))
        y.extend(m(y[-1]) for m in [self.cv2, self.cv3])
        return self.cv4(torch.cat(y, 1))

RepNCSPELAN4 的子模块 RepNCSP

1、RepNCSP 模块的输入参数为（c1, c2, n=1, shortcut=True, g=1, e=0.5）

2、RepNCSP模块模块代码

class RepNCSP(nn.Module):
    # CSP Bottleneck with 3 convolutions
    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c1, c_, 1, 1)
        self.cv3 = Conv(2 * c_, c2, 1)  # optional act=FReLU(c2)
        self.m = nn.Sequential(*(RepNBottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))

    def forward(self, x):
        return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), 1))

RepNCSP中的子模块RepNBottleneck

模块代码

class RepNBottleneck(nn.Module):
    # Standard bottleneck
    def __init__(self, c1, c2, shortcut=True, g=1, k=(3, 3), e=0.5):  # ch_in, ch_out, shortcut, kernels, groups, expand
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = RepConvN(c1, c_, k[0], 1)
        self.cv2 = Conv(c_, c2, k[1], 1, g=g)
        self.add = shortcut and c1 == c2

    def forward(self, x):
        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))

RepNBottleneck中的子模块RepConvN

模块代码

class RepConvN(nn.Module):
    """RepConv is a basic rep-style block, including training and deploy status
    This code is based on https://github.com/DingXiaoH/RepVGG/blob/main/repvgg.py
    """
    default_act = nn.SiLU()  # default activation

    def __init__(self, c1, c2, k=3, s=1, p=1, g=1, d=1, act=True, bn=False, deploy=False):
        super().__init__()
        assert k == 3 and p == 1
        self.g = g
        self.c1 = c1
        self.c2 = c2
        self.act = self.default_act if act is True else act if isinstance(act, nn.Module) else nn.Identity()

        self.bn = None
        self.conv1 = Conv(c1, c2, k, s, p=p, g=g, act=False)
        self.conv2 = Conv(c1, c2, 1, s, p=(p - k // 2), g=g, act=False)

    def forward_fuse(self, x):
        """Forward process"""
        return self.act(self.conv(x))

    def forward(self, x):
        """Forward process"""
        id_out = 0 if self.bn is None else self.bn(x)
        return self.act(self.conv1(x) + self.conv2(x) + id_out)

    def get_equivalent_kernel_bias(self):
        kernel3x3, bias3x3 = self._fuse_bn_tensor(self.conv1)
        kernel1x1, bias1x1 = self._fuse_bn_tensor(self.conv2)
        kernelid, biasid = self._fuse_bn_tensor(self.bn)
        return kernel3x3 + self._pad_1x1_to_3x3_tensor(kernel1x1) + kernelid, bias3x3 + bias1x1 + biasid

    def _avg_to_3x3_tensor(self, avgp):
        channels = self.c1
        groups = self.g
        kernel_size = avgp.kernel_size
        input_dim = channels // groups
        k = torch.zeros((channels, input_dim, kernel_size, kernel_size))
        k[np.arange(channels), np.tile(np.arange(input_dim), groups), :, :] = 1.0 / kernel_size ** 2
        return k

    def _pad_1x1_to_3x3_tensor(self, kernel1x1):
        if kernel1x1 is None:
            return 0
        else:
            return torch.nn.functional.pad(kernel1x1, [1, 1, 1, 1])

    def _fuse_bn_tensor(self, branch):
        if branch is None:
            return 0, 0
        if isinstance(branch, Conv):
            kernel = branch.conv.weight
            running_mean = branch.bn.running_mean
            running_var = branch.bn.running_var
            gamma = branch.bn.weight
            beta = branch.bn.bias
            eps = branch.bn.eps
        elif isinstance(branch, nn.BatchNorm2d):
            if not hasattr(self, 'id_tensor'):
                input_dim = self.c1 // self.g
                kernel_value = np.zeros((self.c1, input_dim, 3, 3), dtype=np.float32)
                for i in range(self.c1):
                    kernel_value[i, i % input_dim, 1, 1] = 1
                self.id_tensor = torch.from_numpy(kernel_value).to(branch.weight.device)
            kernel = self.id_tensor
            running_mean = branch.running_mean
            running_var = branch.running_var
            gamma = branch.weight
            beta = branch.bias
            eps = branch.eps
        std = (running_var + eps).sqrt()
        t = (gamma / std).reshape(-1, 1, 1, 1)
        return kernel * t, beta - running_mean * gamma / std

    def fuse_convs(self):
        if hasattr(self, 'conv'):
            return
        kernel, bias = self.get_equivalent_kernel_bias()
        self.conv = nn.Conv2d(in_channels=self.conv1.conv.in_channels,
                              out_channels=self.conv1.conv.out_channels,
                              kernel_size=self.conv1.conv.kernel_size,
                              stride=self.conv1.conv.stride,
                              padding=self.conv1.conv.padding,
                              dilation=self.conv1.conv.dilation,
                              groups=self.conv1.conv.groups,
                              bias=True).requires_grad_(False)
        self.conv.weight.data = kernel
        self.conv.bias.data = bias
        for para in self.parameters():
            para.detach_()
        self.__delattr__('conv1')
        self.__delattr__('conv2')
        if hasattr(self, 'nm'):
            self.__delattr__('nm')
        if hasattr(self, 'bn'):
            self.__delattr__('bn')
        if hasattr(self, 'id_tensor'):
            self.__delattr__('id_tensor')

ADown

backbone中的ADown是降低feature map的尺寸

模块代码

class ADown(nn.Module):
    def __init__(self, c1, c2):  # ch_in, ch_out, shortcut, kernels, groups, expand
        super().__init__()
        self.c = c2 // 2
        self.cv1 = Conv(c1 // 2, self.c, 3, 2, 1)
        self.cv2 = Conv(c1 // 2, self.c, 1, 1, 0)

    def forward(self, x):
        x = torch.nn.functional.avg_pool2d(x, 2, 1, 0, False, True)
        x1,x2 = x.chunk(2, 1)
        x1 = self.cv1(x1)
        x2 = torch.nn.functional.max_pool2d(x2, 3, 2, 1)
        x2 = self.cv2(x2)
        return torch.cat((x1, x2), 1)