博客

一、C3模块

在原版YOLOv5网络中，C3模块的结构如图1-1所示，C3结构中的ConvBNSiLU和BottleNeck的结构如图1-2所示：

Conv、BottleNeck、C3的代码如下：

class Conv(nn.Module):
    # Standard convolution  通用卷积模块,包括1卷积1BN1激活,激活默认SiLU,可用变量指定,不激活时用nn.Identity()占位,直接返回输入
    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
        super(Conv, self).__init__()
        self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False)
        self.bn = nn.BatchNorm2d(c2)
        self.act = nn.SiLU() if act is True else (act if isinstance(act, nn.Module) else nn.Identity())

    def forward(self, x):
        return self.act(self.bn(self.conv(x)))

    def fuseforward(self, x):
        return self.act(self.conv(x))


class Bottleneck(nn.Module):
    # Standard bottleneck 残差块
    def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, shortcut, groups, expansion
        super(Bottleneck, self).__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c_, c2, 3, 1, g=g)
        self.add = shortcut and c1 == c2

    def forward(self, x):  # 如果shortcut并且输入输出通道相同则跳层相加
        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))

class C3(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(C3, self).__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)  # act=FReLU(c2)
        self.m = nn.Sequential(*[Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)])  # n个残差组件(Bottleneck)
        # self.m = nn.Sequential(*[CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)])

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

二、轻量化C3模块

轻量化C3的改进思路是将原C3模块中使用的普通卷积，全部替换为深度可分离卷积，其余结构不变，改进后的DP_Conv、DP_BottleNeck、DP_C3的代码如下：

class DP_Conv(nn.Module):
    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
        super(DP_Conv, self).__init__()
        self.conv = nn.Conv2d(c1, c1, kernel_size=3, stride=1, padding=1, groups=c1)
        self.conv = nn.Conv2d(c1, c2, kernel_size=1, stride=s)
        self.bn = nn.BatchNorm2d(c2)
        self.act = nn.SiLU() if act is True else (act if isinstance(act, nn.Module) else nn.Identity())

    def forward(self, x):
        return self.act(self.bn(self.conv(x)))

    def fuseforward(self, x):
        return self.act(self.conv(x))

class DP_Bottleneck(nn.Module):
    def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, shortcut, groups, expansion
        super(DP_Bottleneck, self).__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = DP_Conv(c1, c_, 1)
        self.cv2 = DP_Conv(c_, c2, 1)
        self.add = shortcut and c1 == c2

    def forward(self, x):  # 如果shortcut并且输入输出通道相同则跳层相加
        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))

class DP_C3(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(DP_C3, self).__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = DP_Conv(c1, c_, 1)
        self.cv2 = DP_Conv(c1, c_, 1)
        self.cv3 = DP_Conv(2 * c_, c2, 1)  # act=FReLU(c2)
        self.m = nn.Sequential(*[DP_Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)])  # n个残差组件(Bottleneck)
        # self.m = nn.Sequential(*[CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)])

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

修改后的网络结构如下：

# parameters
nc: 20  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple

# anchors
anchors:
  - [10,13, 16,30, 33,23]  # P3/8
  - [30,61, 62,45, 59,119]  # P4/16
  - [116,90, 156,198, 373,326]  # P5/32

# YOLOv5 backbone
backbone:
  # [from, number, module, args]
  [[-1, 1, DP_Conv, [64, 2]],  # 0-P1/2
   [-1, 1, DP_Conv, [128, 2]],  # 1-P2/4
   [-1, 3, DP_C3, [128]],
   [-1, 1, DP_Conv, [256, 2]],  # 3-P3/8
   [-1, 9, DP_C3, [256]],
   [-1, 1, DP_Conv, [512, 2]],  # 5-P4/16
   [-1, 9, DP_C3, [512]],
   [-1, 1, DP_Conv, [1024, 2]],  # 7-P5/32
   [-1, 3, DP_C3, [1024]],
   [-1, 1, SPPF, [1024, 5]], # 9
  ]

# YOLOv5 head
head:
  [[-1, 1, DP_Conv, [512, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [[-1, 6], 1, Concat, [1]],  # cat backbone P4  # PANet是add, yolov5是concat
   [-1, 3, C3, [512, False]],  # 13

   [-1, 1, DP_Conv, [256, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)

   [-1, 1, DP_Conv, [256, 2]],
   [[-1, 14], 1, Concat, [1]],  # cat head P4
   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)

   [-1, 1, DP_Conv, [512, 2]],
   [[-1, 10], 1, Concat, [1]],  # cat head P5
   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)

   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)  必须在最后一层, 原代码很多默认了Detect是最后, 并没有全改
  ]