yolov5 增加檢測層

參考1：https://blog.csdn.net/weixin_41868104/article/details/111596851
參考2：https://blog.csdn.net/jacke121/article/details/118714043

原模型結構

原配置文件

# YOLOv5  by Ultralytics, GPL-3.0 license

# Parameters
nc: 80  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple
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, Focus, [64, 3]],  # 0-P1/2
   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
   [-1, 3, C3, [128]],
   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
   [-1, 9, C3, [256]],
   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
   [-1, 9, C3, [512]],
   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
   [-1, 1, SPP, [1024, [5, 9, 13]]],
   [-1, 3, C3, [1024, False]],  # 9
  ]

# YOLOv5 head
head:
  [[-1, 1, Conv, [512, 1, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
   [-1, 3, C3, [512, False]],  # 13

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

   [-1, 1, 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)
  ]

新模型結構

新配置文件

# YOLOv5  by Ultralytics, GPL-3.0 license

# Parameters
nc: 80  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple
anchors:
  - [ 5,6,8,14,15,11 ]
  - [ 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, Focus, [ 64, 3 ] ],  # 0-P1/2
    [ -1, 1, Conv, [ 128, 3, 2 ] ],  # 1-P2/4
    [ -1, 3, C3, [ 128 ] ],   # 2
    [ -1, 1, Conv, [ 256, 3, 2 ] ],  # 3-P3/8
    [ -1, 9, C3, [ 256 ] ],  # 4
    [ -1, 1, Conv, [ 512, 3, 2 ] ],  # 5-P4/16
    [ -1, 9, C3, [ 512 ] ],   # 6
    [ -1, 1, Conv, [ 1024, 3, 2 ] ],  # 7-P5/32
    [ -1, 1, SPP, [ 1024, [ 5, 9, 13 ] ] ],
    [ -1, 3, C3, [ 1024, False ] ],  # 9
  ]

# YOLOv5 head
head:
  [ [ -1, 1, Conv, [ 512, 1, 1 ] ],          # 10
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],  # 11
    [ [ -1, 6 ], 1, Concat, [ 1 ] ],  # cat backbone P4 #12
    [ -1, 3, C3, [ 512, False ] ],  # 13

    [ -1, 1, Conv, [ 512, 1, 1 ] ], # 14
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], # 15
    [ [ -1, 4 ], 1, Concat, [ 1 ] ],  # cat backbone P3 16
    [ -1, 3, C3, [ 512, False ] ],  # 17

    [ -1, 1, Conv, [ 256, 1, 1 ] ], # 18
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], #19
    [ [ -1, 2 ], 1, Concat, [ 1 ] ],  # cat backbone P3 # 20
    [ -1, 3, C3, [ 256, False ] ],  # 21 (P3/8-small)  21

    [ -1, 1, Conv, [ 256, 3, 2 ] ],   #22
    [ [ -1, 18 ], 1, Concat, [ 1 ] ],  # cat backbone P3 # 23
    [ -1, 3, C3, [ 256, False ] ],  # 24 (P3/8-small)

    [ -1, 1, Conv, [ 256, 3, 2 ] ],
    [ [ -1, 14 ], 1, Concat, [ 1 ] ],  # cat head P4
    [ -1, 3, C3, [ 512, False ] ],  # 27 (P4/16-medium)

    [ -1, 1, Conv, [ 512, 3, 2 ] ],
    [ [ -1, 10 ], 1, Concat, [ 1 ] ],  # cat head P5
    [ -1, 3, C3, [ 1024, False ] ],  # 30 (P5/32-large)
    [ [ 21,24,27,30 ], 1, Detect, [ nc, anchors ] ],  # Detect(P3, P4, P5)
  ]

主要修改點：
1、 添加針對小目標的 Anchor參數
2、 從骨干網絡的第2層就開始增強特征
3、 添加針對第2層的檢測頭部

按照原作者(參考1)上解釋：
在第17層后，繼續對特征圖進行上采樣等處理，使得特征圖繼續擴大，同時在第20層時，將獲取到的大小為160X160的特征圖與骨干網絡中第2層特征圖進行concat融合，以此獲取更大的特征圖進行小目標檢測。

結果

yolov5s1 原生的yolov5s模型
yolov5s2 增加檢測頭的yolov5s模型

由於yolov5s2並不能完全遷移之前的權重數據，需要重新訓練模型，其中精確度並不一定比yolov5s高。

以下結果 僅僅代表個人實驗

	yolov5s1	yolov5s2	yolov5m
layers	283	341	391
parameter	7063542	7724296	21056406
gpu_mem	10G	11.4G	12.6G
weights	14M	15M	41M
GFLOPs	16.4	27.5	50.4
time	0.0017	0.0023	0.034
max_ap05	0.8856	0.89846	0.9016

附加

此外論文  《TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone》  

針對高空無人機對地面目標的檢測 也采用了該四個檢測頭結構
論文鏈接： https://arxiv.org/pdf/2108.11539.pdf
論文解讀： https://zhuanlan.zhihu.com/p/410166419