YOLO算法改进6【中阶改进篇】：depthwise separable convolution轻量化C3

常规卷积操作

对于一张5×5像素、三通道（shape为5×5×3），经过3×3卷积核的卷积层（假设输出通道数为4，则卷积核shape为3×3×3×4，最终输出4个Feature Map，如果有same padding则尺寸与输入层相同（5×5），如果没有则为尺寸变为3×3

深度可分离卷积

• 逐通道卷积Depthwise Convolution

Depthwise Convolution的一个卷积核负责一个通道，一个通道只被一个卷积核卷积。

一张5×5像素、三通道彩色输入图片（shape为5×5×3），Depthwise Convolution首先经过第一次卷积运算，DW完全是在二维平面内进行。卷积核的数量与上一层的通道数相同（通道和卷积核一一对应）。所以一个三通道的图像经过运算后生成了3个Feature map(如果有same padding则尺寸与输入层相同为5×5)，如下图所示。
Depthwise Convolution完成后的Feature map数量与输入层的通道数相同，无法扩展Feature map。而且这种运算对输入层的每个通道独立进行卷积运算，没有有效的利用不同通道在相同空间位置上的feature信息。因此需要Pointwise Convolution来将这些Feature map进行组合生成新的Feature map

• 逐点卷积Pointwise Convolution

Pointwise Convolution的运算与常规卷积运算非常相似，它的卷积核的尺寸为 1×1×M，M为上一层的通道数。所以这里的卷积运算会将上一步的map在深度方向上进行加权组合，生成新的Feature map。有几个卷积核就有几个输出Feature map

经过Pointwise Convolution之后，同样输出了4张Feature map，与常规卷积的输出维度相同

YOLOV5s中Conv、BottleNeck、C3的代码如下:

原始common.py配置

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))

1.轻量化C3模块

在models/common.py文件中按以下思路修改代码：
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.conv1 = nn.Conv2d(c1, c1, kernel_size=3, stride=1, padding=1, groups=c1)
        self.conv2 = 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.conv2(self.conv1(x))))

    def fuseforward(self, x):
        return self.act(self.conv2(self.conv1(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))

2.添加DP_C3.yaml文件
添加至/models/文件中

# parameters
nc: 80  # 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, 6, 2, 2]],  # 0-P1/2
   [-1, 1, DP_Conv, [128, 3, 2]],  # 1-P2/4
   [-1, 3, DP_C3, [128]],
   [-1, 1, DP_Conv, [256, 3, 2]],  # 3-P3/8
   [-1, 9, DP_C3, [256]],
   [-1, 1, DP_Conv, [512, 3, 2]],  # 5-P4/16
   [-1, 9, DP_C3, [512]],
   [-1, 1, DP_Conv, [1024, 3, 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, 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, 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, 3,2]],
   [[-1, 14], 1, Concat, [1]],  # cat head P4
   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)

   [-1, 1, DP_Conv, [512, 3, 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
  ]

3.yolo.py配置
找到 models/yolo.py 文件中 parse_model() 类 for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']):，在列表中添加DP_Conv、DP_BottleNeck、DP_C3，这样可以获得我们要传入的参数。

 if m in {
                Conv, GhostConv, Bottleneck, GhostBottleneck, SPP, SPPF, DWConv, MixConv2d, Focus, CrossConv,
                BottleneckCSP, C3, C3TR, C3SPP, C3Ghost, nn.ConvTranspose2d, DWConvTranspose2d, C3x,Attention, CondConv, DP_Conv, DP_BottleNeck, DP_C3}:
            c1, c2 = ch[f], args[0]
            if c2 != no:  # if not output
                c2 = make_divisible(c2 * gw, 8)

            args = [c1, c2, *args[1:]]
            if m in {BottleneckCSP, C3, C3TR, C3Ghost, C3x, DP_C3}:
                args.insert(2, n)  # number of repeats
                n = 1
        elif m is nn.BatchNorm2d:
            args = [ch[f]]
        elif m is Concat:
            c2 = sum(ch[x] for x in f)
        # TODO: channel, gw, gd

4.训练模型

python train.py --cfg DP_C3.yaml