YOLOv7源码解析

YAML文件

以yolov7 cfg/yolov7-w6-pose.yaml为例：

# parameters

nc: 1  # number of classes
nkpt: 4 # number of key points
depth_multiple: 1.0  # model depth multiple
width_multiple: 1.0  # layer channel multiple
dw_conv_kpt: True

anchors:
  - [ 19,27,  44,40,  38,94 ]  # P3/8
  - [ 96,68,  86,152,  180,137 ]  # P4/16
  - [ 140,301,  303,264,  238,542 ]  # P5/32
  - [ 436,615,  739,380,  925,792 ]  # P6/64

# yolov7 backbone
backbone:
  [[-1, 1, ReOrg, []],  # 0
   [-1, 1, Conv, [64, 3, 1]],  # 1-P1/2
   
   [-1, 1, Conv, [128, 3, 2]],  # 2-P2/4
   [-1, 1, Conv, [64, 1, 1]],
   [-2, 1, Conv, [64, 1, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [[-1, -3, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [128, 1, 1]],  # 10
         
   [-1, 1, Conv, [256, 3, 2]],  # 11-P3/8
   [-1, 1, Conv, [128, 1, 1]],
   [-2, 1, Conv, [128, 1, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [[-1, -3, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [256, 1, 1]],  # 19
         
   [-1, 1, Conv, [512, 3, 2]],  # 20-P4/16
   [-1, 1, Conv, [256, 1, 1]],
   [-2, 1, Conv, [256, 1, 1]],
   [-1, 1, Conv, [256, 3, 1]],
   [-1, 1, Conv, [256, 3, 1]],
   [-1, 1, Conv, [256, 3, 1]],
   [-1, 1, Conv, [256, 3, 1]],
   [[-1, -3, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [512, 1, 1]],  # 28
         
   [-1, 1, Conv, [768, 3, 2]],  # 29-P5/32
   [-1, 1, Conv, [384, 1, 1]],
   [-2, 1, Conv, [384, 1, 1]],
   [-1, 1, Conv, [384, 3, 1]],
   [-1, 1, Conv, [384, 3, 1]],
   [-1, 1, Conv, [384, 3, 1]],
   [-1, 1, Conv, [384, 3, 1]],
   [[-1, -3, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [768, 1, 1]],  # 37
         
   [-1, 1, Conv, [1024, 3, 2]],  # 38-P6/64
   [-1, 1, Conv, [512, 1, 1]],
   [-2, 1, Conv, [512, 1, 1]],
   [-1, 1, Conv, [512, 3, 1]],
   [-1, 1, Conv, [512, 3, 1]],
   [-1, 1, Conv, [512, 3, 1]],
   [-1, 1, Conv, [512, 3, 1]],
   [[-1, -3, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [1024, 1, 1]],  # 46
  ]

# yolov7 head
head:
  [[-1, 1, SPPCSPC, [512]], # 47
  
   [-1, 1, Conv, [384, 1, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [37, 1, Conv, [384, 1, 1]], # route backbone P5
   [[-1, -2], 1, Concat, [1]],
   
   [-1, 1, Conv, [384, 1, 1]],
   [-2, 1, Conv, [384, 1, 1]],
   [-1, 1, Conv, [192, 3, 1]],
   [-1, 1, Conv, [192, 3, 1]],
   [-1, 1, Conv, [192, 3, 1]],
   [-1, 1, Conv, [192, 3, 1]],
   [[-1, -2, -3, -4, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [384, 1, 1]], # 59
  
   [-1, 1, Conv, [256, 1, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [28, 1, Conv, [256, 1, 1]], # route backbone P4
   [[-1, -2], 1, Concat, [1]],
   
   [-1, 1, Conv, [256, 1, 1]],
   [-2, 1, Conv, [256, 1, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [[-1, -2, -3, -4, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [256, 1, 1]], # 71
   
   [-1, 1, Conv, [128, 1, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [19, 1, Conv, [128, 1, 1]], # route backbone P3
   [[-1, -2], 1, Concat, [1]],
   
   [-1, 1, Conv, [128, 1, 1]],
   [-2, 1, Conv, [128, 1, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [[-1, -2, -3, -4, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [128, 1, 1]], # 83
      
   [-1, 1, Conv, [256, 3, 2]],
   [[-1, 71], 1, Concat, [1]],  # cat
   
   [-1, 1, Conv, [256, 1, 1]],
   [-2, 1, Conv, [256, 1, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [-1, 1, Conv, [128, 3, 1]],
   [[-1, -2, -3, -4, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [256, 1, 1]], # 93
      
   [-1, 1, Conv, [384, 3, 2]],
   [[-1, 59], 1, Concat, [1]],  # cat
   
   [-1, 1, Conv, [384, 1, 1]],
   [-2, 1, Conv, [384, 1, 1]],
   [-1, 1, Conv, [192, 3, 1]],
   [-1, 1, Conv, [192, 3, 1]],
   [-1, 1, Conv, [192, 3, 1]],
   [-1, 1, Conv, [192, 3, 1]],
   [[-1, -2, -3, -4, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [384, 1, 1]], # 103
      
   [-1, 1, Conv, [512, 3, 2]],
   [[-1, 47], 1, Concat, [1]],  # cat
   
   [-1, 1, Conv, [512, 1, 1]],
   [-2, 1, Conv, [512, 1, 1]],
   [-1, 1, Conv, [256, 3, 1]],
   [-1, 1, Conv, [256, 3, 1]],
   [-1, 1, Conv, [256, 3, 1]],
   [-1, 1, Conv, [256, 3, 1]],
   [[-1, -2, -3, -4, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [512, 1, 1]], # 113
   
   [83, 1, Conv, [256, 3, 1]],
   [93, 1, Conv, [512, 3, 1]],
   [103, 1, Conv, [768, 3, 1]],
   [113, 1, Conv, [1024, 3, 1]],

   [[114,115,116,117], 1, IKeypoint, [nc, anchors, nkpt]],   # Detect(P3, P4, P5, P6)
  ]

其中头部部分：

nc: 1  # number of classes
nkpt: 4 # number of key points
depth_multiple: 1.0  # model depth multiple
width_multiple: 1.0  # layer channel multiple
dw_conv_kpt: True

• nc：表示任务类别个数。例如做人、车、狗检测，此时nc=3
• nkpt：表示关键点的数量。如做人的17个关键点检测，此时nkpt=17
• dept_multiple：表示模型的深度
• width_multiple：表示模型的宽度

其中backbone部分：

[-1, 1, Conv, [128, 3, 2]],  # 2-P2/4
   [-1, 1, Conv, [64, 1, 1]],
   [-2, 1, Conv, [64, 1, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [-1, 1, Conv, [64, 3, 1]],
   [[-1, -3, -5, -6], 1, Concat, [1]],
   [-1, 1, Conv, [128, 1, 1]],  # 10

这一部分从P2（第2层开始)，向下一次为第3层，第4层，…，第10层。
其中

• 第一个参数：-1表示当前层的输入来自上一层，若是-2表示当前层（i）的输入来自第（i-2）层。
• 第二个参数：1表示这个参数乘上模型深度的超参数，就可以控制模型的深度
• 第三个参数：Conv表示具体的网络层
• 第四个参数：输出通道、卷积核等大小
  • Conv：输出通道、卷积核大小、步长
  • SPP：输出通道、卷积核大小
  • Foucs：输出通道、卷积核大小
  • BottleckCSP：输出通道、是否启用shortcut
  • Concat：拼接维度
  • Detect：类别个数、anchors

其中尾部部分：

 [[114,115,116,117], 1, IKeypoint, [nc, anchors, nkpt]],   # Detect(P3, P4, P5, P6)

• [114,115,116,117]：表示检测层的特征图来源，此时检测层层数为4
• IKeypoint：关键点检测
• [nc, anchors, nkpt]：
  • nc ：表示类别
  • anchor：表示anchor
  • nkpt：表示关键点数量