论文题目：Slim-neck by GSConv: A better design paradigm of detector architectures for autonomous vehicles

论文：https://arxiv.org/abs/2206.02424

代码：https://github.com/AlanLi1997/Slim-neck-by-GSConv

直接步入正题~~~

目标：为YOLOv5模型构建一个简单高效的Neck模块。考虑了卷积方法、特征融合结构、计算效率、计算成本效益等诸多因素。

一、GSConv

class GSConv(nn.Module):
    # GSConv https://github.com/AlanLi1997/slim-neck-by-gsconv
    def __init__(self, c1, c2, k=1, s=1, g=1, act=True):
        super().__init__()
        c_ = c2 // 2
        self.cv1 = Conv(c1, c_, k, s, None, g, act)
        self.cv2 = Conv(c_, c_, 5, 1, None, c_, act)
    def forward(self, x):
        x1 = self.cv1(x)
        x2 = torch.cat((x1, self.cv2(x1)), 1)
        # shuffle
        b, n, h, w = x2.data.size()
        b_n = b * n // 2
        y = x2.reshape(b_n, 2, h * w)
        y = y.permute(1, 0, 2)
        y = y.reshape(2, -1, n // 2, h, w)
        return torch.cat((y[0], y[1]), 1)

将YOLOv5s.yaml的Neck模块中的Conv换成GSConv

1、将GSConv代码加入common.py文件中

2、找到yolo.py文件里的parse_model函数，将类名加入进去

3、修改配置文件，将YOLOv5s.yaml的Neck模块中的Conv换成GSConv 

~~~此处有一个疑问，官方给出的GSConv代码中为什么没用DWConv呢？希望知道的朋友在评论区指点一下~~~

二、GSConv+Slim Neck

1、GSBottleneck

class GSBottleneck(nn.Module):
    # GS Bottleneck https://github.com/AlanLi1997/slim-neck-by-gsconv
    def __init__(self, c1, c2, k=3, s=1):
        super().__init__()
        c_ = c2 // 2
        # for lighting
        self.conv_lighting = nn.Sequential(
            GSConv(c1, c_, 1, 1),
            GSConv(c_, c2, 1, 1, act=False))
        # for receptive field
        self.conv = nn.Sequential(
            GSConv(c1, c_, 3, 1),
            GSConv(c_, c2, 3, 1, act=False))
        self.shortcut = Conv(c1, c2, 3, 1, act=False)

    def forward(self, x):
        return self.conv_lighting(x) + self.shortcut(x)

2、VoVGSCSP

class VoVGSCSP(nn.Module):
    # VoV-GSCSP https://github.com/AlanLi1997/slim-neck-by-gsconv
    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        super().__init__()
        c_ = int(c2 * e)
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(2 * c_, c2, 1)
        self.m = nn.Sequential(*(GSBottleneck(c_, c_) for _ in range(n)))
    def forward(self, x):
        x1 = self.cv1(x)
        return self.cv2(torch.cat((self.m(x1), x1), dim=1))

将YOLOv5s.yaml的Neck模块中的Conv换成GSConv，C3模块换为VoVGSCSP模块

1、将以下代码加入common.py文件中

class GSConv(nn.Module):
    # GSConv https://github.com/AlanLi1997/slim-neck-by-gsconv
    def __init__(self, c1, c2, k=1, s=1, g=1, act=True):
        super().__init__()
        c_ = c2 // 2
        self.cv1 = Conv(c1, c_, k, s, None, g, act)
        self.cv2 = Conv(c_, c_, 5, 1, None, c_, act)
    def forward(self, x):
        x1 = self.cv1(x)
        x2 = torch.cat((x1, self.cv2(x1)), 1)
        # shuffle
        b, n, h, w = x2.data.size()
        b_n = b * n // 2
        y = x2.reshape(b_n, 2, h * w)
        y = y.permute(1, 0, 2)
        y = y.reshape(2, -1, n // 2, h, w)
        return torch.cat((y[0], y[1]), 1)

class GSBottleneck(nn.Module):
    # GS Bottleneck https://github.com/AlanLi1997/slim-neck-by-gsconv
    def __init__(self, c1, c2, k=3, s=1):
        super().__init__()
        c_ = c2 // 2
        # for lighting
        self.conv_lighting = nn.Sequential(
            GSConv(c1, c_, 1, 1),
            GSConv(c_, c2, 1, 1, act=False))
        # for receptive field
        self.conv = nn.Sequential(
            GSConv(c1, c_, 3, 1),
            GSConv(c_, c2, 3, 1, act=False))
        self.shortcut = nn.Identity()
    def forward(self, x):
        return self.conv_lighting(x)


class VoVGSCSP(nn.Module):
    # VoV-GSCSP https://github.com/AlanLi1997/slim-neck-by-gsconv
    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        super().__init__()
        c_ = int(c2 * e)
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(2 * c_, c2, 1)
        self.m = nn.Sequential(*(GSBottleneck(c_, c_) for _ in range(n)))
    def forward(self, x):
        x1 = self.cv1(x)
        return self.cv2(torch.cat((self.m(x1), x1), dim=1))

2、找到yolo.py文件里的parse_model函数，将类名加入进去，注意有两处需要添加的地方

3、修改配置文件，将YOLOv5s.yaml的Neck模块中的Conv换成GSConv ，C3模块换为VoVGSCSP

 

Appendix

下图是原论文中给出的结构图，个人对照源码后觉得这里多画了一个GSConv模块(红色框里所示)，如果有知道的大佬望在评论区指点一下。