devbox
Guff_9hys
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

【YOLOv5/v7改进系列】改进池化层为RFB

作者：Guff_9hys | 2024-07-09 10:51:05

赞

踩

【YOLOv5/v7改进系列】改进池化层为RFB
一、导言

论文 "Receptive Field Block Net for Accurate and Fast Object Detection" 中提出的 RFB (Receptive Field Block) 模块旨在模仿人类视觉系统中的感受野结构，以增强深度学习模型对不同尺度和位置的目标检测能力。下面总结了RFB模块的主要优点和潜在局限性。

优点:

  1. 增强特征区分性和鲁棒性: RFB模块通过模拟人类视觉系统中感受野的大小和偏心率之间的关系，增强了轻量级CNN网络的特征表示，从而提高了特征的区分性和鲁棒性。

  2. 多尺度信息整合: RFB模块利用多个分支，每个分支有不同的卷积核大小和空洞卷积层，这有助于捕捉图像中不同尺度的信息，类似于人类视觉系统中感受野的多样性。

  3. 实时性能: 实验表明，将RFB模块添加到SSD之上构建的RFB Net不仅保持了实时处理速度，而且在Pascal VOC和MS COCO数据集上达到了与高级深层检测器相当的性能。

  4. 通用性: RFB模块可以很容易地链接到不同的轻量级网络架构上，如MobileNet，这证明了它的通用性和适应性。

  5. 从零开始训练的能力: RFB模块还显示出即使在没有预训练的情况下也能有效训练物体检测器的能力，这在某些情况下可能是一个优势。

局限性:

  1. 参数增加: 尽管RFB模块可以提高精度，但它也增加了额外的层和参数，这可能会影响计算效率和资源需求，尤其是在低端设备上。

  2. 特定设计的局限性: RFB模块的设计是手工制作的，这意味着它可能不如自适应机制灵活，例如可变形卷积神经网络中使用的那些，后者可以根据对象的尺度和形状调整感受野的分布。

  3. 对感受野的简化假设: RFB模块通过固定大小和偏心率的感受野来近似人类视觉系统，这可能忽略了更复杂的人类视觉处理机制，如动态调整或非均匀采样。

  4. 依赖于基础网络: RFB模块的性能很大程度上依赖于底层的基础网络，这意味着在不同的网络架构上的效果可能会有所不同。

尽管存在这些局限性，RFB模块在物体检测任务中表现出了显著的改进，并且在保持实时处理速度的同时，提供了一种增强轻量级模型检测精度的有效方法。

二、准备工作

首先在YOLOv5/v7的models文件夹下新建文件morepooling.py，导入如下代码

  1. from models.common import *
  2.  
  3. # RFB
  4. # https://arxiv.org/pdf/1711.07767
  5. class BasicConv(nn.Module):
  6.  
  7.     def __init__(self, in_planes, out_planes, kernel_size, stride=1, padding=0, dilation=1, groups=1, relu=True,
  8.                  bn=True):
  9.         super(BasicConv, self).__init__()
  10.         self.out_channels = out_planes
  11.         if bn:
  12.             self.conv = nn.Conv2d(in_planes, out_planes, kernel_size=kernel_size, stride=stride, padding=padding,
  13.                                   dilation=dilation, groups=groups, bias=False)
  14.             self.bn = nn.BatchNorm2d(out_planes, eps=1e-5, momentum=0.01, affine=True)
  15.             self.relu = nn.ReLU(inplace=True) if relu else None
  16.         else:
  17.             self.conv = nn.Conv2d(in_planes, out_planes, kernel_size=kernel_size, stride=stride, padding=padding,
  18.                                   dilation=dilation, groups=groups, bias=True)
  19.             self.bn = None
  20.             self.relu = nn.ReLU(inplace=True) if relu else None
  21.  
  22.     def forward(self, x):
  23.         x = self.conv(x)
  24.         if self.bn is not None:
  25.             x = self.bn(x)
  26.         if self.relu is not None:
  27.             x = self.relu(x)
  28.         return x
  29.  
  30.  
  31. class BasicRFB(nn.Module):
  32.  
  33.     def __init__(self, in_planes, out_planes, stride=1, scale=0.1, map_reduce=8, vision=1, groups=1):
  34.         super(BasicRFB, self).__init__()
  35.         self.scale = scale
  36.         self.out_channels = out_planes
  37.         inter_planes = in_planes // map_reduce
  38.  
  39.         self.branch0 = nn.Sequential(
  40.             BasicConv(in_planes, inter_planes, kernel_size=1, stride=1, groups=groups, relu=False),
  41.             BasicConv(inter_planes, 2 * inter_planes, kernel_size=(3, 3), stride=stride, padding=(1, 1), groups=groups),
  42.             BasicConv(2 * inter_planes, 2 * inter_planes, kernel_size=3, stride=1, padding=vision, dilation=vision,
  43.                       relu=False, groups=groups)
  44.         )
  45.         self.branch1 = nn.Sequential(
  46.             BasicConv(in_planes, inter_planes, kernel_size=1, stride=1, groups=groups, relu=False),
  47.             BasicConv(inter_planes, 2 * inter_planes, kernel_size=(3, 3), stride=stride, padding=(1, 1), groups=groups),
  48.             BasicConv(2 * inter_planes, 2 * inter_planes, kernel_size=3, stride=1, padding=vision + 2,
  49.                       dilation=vision + 2, relu=False, groups=groups)
  50.         )
  51.         self.branch2 = nn.Sequential(
  52.             BasicConv(in_planes, inter_planes, kernel_size=1, stride=1, groups=groups, relu=False),
  53.             BasicConv(inter_planes, (inter_planes // 2) * 3, kernel_size=3, stride=1, padding=1, groups=groups),
  54.             BasicConv((inter_planes // 2) * 3, 2 * inter_planes, kernel_size=3, stride=stride, padding=1,
  55.                       groups=groups),
  56.             BasicConv(2 * inter_planes, 2 * inter_planes, kernel_size=3, stride=1, padding=vision + 4,
  57.                       dilation=vision + 4, relu=False, groups=groups)
  58.         )
  59.  
  60.         self.ConvLinear = BasicConv(6 * inter_planes, out_planes, kernel_size=1, stride=1, relu=False)
  61.         self.shortcut = BasicConv(in_planes, out_planes, kernel_size=1, stride=stride, relu=False)
  62.         self.relu = nn.ReLU(inplace=False)
  63.  
  64.     def forward(self, x):
  65.         x0 = self.branch0(x)
  66.         x1 = self.branch1(x)
  67.         x2 = self.branch2(x)
  68.  
  69.         out = torch.cat((x0, x1, x2), 1)
  70.         out = self.ConvLinear(out)
  71.         short = self.shortcut(x)
  72.         out = out * self.scale + short
  73.         out = self.relu(out)
  74.  
  75.         return out

其次在在YOLOv5/v7项目文件下的models/yolo.py中在文件首部添加代码

from models.morepooling import *

并搜索def parse_model(d, ch)

定位到如下行添加以下代码

BasicRFB

三、YOLOv7-tiny改进工作

完成二后，在YOLOv7项目文件下的models文件夹下创建新的文件yolov7-tiny-rfb.yaml，导入如下代码。

  1. # parameters
  2. nc: 80  # number of classes
  3. depth_multiple: 1.0  # model depth multiple
  4. width_multiple: 1.0  # layer channel multiple
  5.  
  6. # anchors
  7. anchors:
  8.   - [10,13, 16,30, 33,23]  # P3/8
  9.   - [30,61, 62,45, 59,119]  # P4/16
  10.   - [116,90, 156,198, 373,326]  # P5/32
  11.  
  12. # yolov7-tiny backbone
  13. backbone:
  14.   # [from, number, module, args] c2, k=1, s=1, p=None, g=1, act=True
  15.   [[-1, 1, Conv, [32, 3, 2, None, 1, nn.LeakyReLU(0.1)]],  # 0-P1/2
  16.   
  17.    [-1, 1, Conv, [64, 3, 2, None, 1, nn.LeakyReLU(0.1)]],  # 1-P2/4
  18.    
  19.    [-1, 1, Conv, [32, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  20.    [-2, 1, Conv, [32, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  21.    [-1, 1, Conv, [32, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  22.    [-1, 1, Conv, [32, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  23.    [[-1, -2, -3, -4], 1, Concat, [1]],
  24.    [-1, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],  # 7
  25.    
  26.    [-1, 1, MP, []],  # 8-P3/8
  27.    [-1, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  28.    [-2, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  29.    [-1, 1, Conv, [64, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  30.    [-1, 1, Conv, [64, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  31.    [[-1, -2, -3, -4], 1, Concat, [1]],
  32.    [-1, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]],  # 14
  33.    
  34.    [-1, 1, MP, []],  # 15-P4/16
  35.    [-1, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  36.    [-2, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  37.    [-1, 1, Conv, [128, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  38.    [-1, 1, Conv, [128, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  39.    [[-1, -2, -3, -4], 1, Concat, [1]],
  40.    [-1, 1, Conv, [256, 1, 1, None, 1, nn.LeakyReLU(0.1)]],  # 21
  41.    
  42.    [-1, 1, MP, []],  # 22-P5/32
  43.    [-1, 1, Conv, [256, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  44.    [-2, 1, Conv, [256, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  45.    [-1, 1, Conv, [256, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  46.    [-1, 1, Conv, [256, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  47.    [[-1, -2, -3, -4], 1, Concat, [1]],
  48.    [-1, 1, Conv, [512, 1, 1, None, 1, nn.LeakyReLU(0.1)]],  # 28
  49.   ]
  50.  
  51. # yolov7-tiny head
  52. head:
  53.   [[-1, 1, BasicRFB, [256]], # 29
  54.   
  55.    [-1, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  56.    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
  57.    [21, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]], # route backbone P4
  58.    [[-1, -2], 1, Concat, [1]],
  59.    
  60.    [-1, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  61.    [-2, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  62.    [-1, 1, Conv, [64, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  63.    [-1, 1, Conv, [64, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  64.    [[-1, -2, -3, -4], 1, Concat, [1]],
  65.    [-1, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]],  # 39
  66.   
  67.    [-1, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  68.    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
  69.    [14, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]], # route backbone P3
  70.    [[-1, -2], 1, Concat, [1]],
  71.    
  72.    [-1, 1, Conv, [32, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  73.    [-2, 1, Conv, [32, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  74.    [-1, 1, Conv, [32, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  75.    [-1, 1, Conv, [32, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  76.    [[-1, -2, -3, -4], 1, Concat, [1]],
  77.    [-1, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],  # 49
  78.    
  79.    [-1, 1, Conv, [128, 3, 2, None, 1, nn.LeakyReLU(0.1)]],
  80.    [[-1, 39], 1, Concat, [1]],
  81.    
  82.    [-1, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  83.    [-2, 1, Conv, [64, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  84.    [-1, 1, Conv, [64, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  85.    [-1, 1, Conv, [64, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  86.    [[-1, -2, -3, -4], 1, Concat, [1]],
  87.    [-1, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]],  # 57
  88.    
  89.    [-1, 1, Conv, [256, 3, 2, None, 1, nn.LeakyReLU(0.1)]],
  90.    [[-1, 29], 1, Concat, [1]],
  91.    
  92.    [-1, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  93.    [-2, 1, Conv, [128, 1, 1, None, 1, nn.LeakyReLU(0.1)]],
  94.    [-1, 1, Conv, [128, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  95.    [-1, 1, Conv, [128, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  96.    [[-1, -2, -3, -4], 1, Concat, [1]],
  97.    [-1, 1, Conv, [256, 1, 1, None, 1, nn.LeakyReLU(0.1)]],  # 65
  98.       
  99.    [49, 1, Conv, [128, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  100.    [57, 1, Conv, [256, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  101.    [65, 1, Conv, [512, 3, 1, None, 1, nn.LeakyReLU(0.1)]],
  102.  
  103.    [[66, 67, 68], 1, IDetect, [nc, anchors]],   # Detect(P3, P4, P5)
  104.   ]
  1.  
  2.                  from  n    params  module                                  arguments                     
  3.   0                -1  1       928  models.common.Conv                      [3, 32, 3, 2, None, 1, LeakyReLU(negative_slope=0.1)]
  4.   1                -1  1     18560  models.common.Conv                      [32, 64, 3, 2, None, 1, LeakyReLU(negative_slope=0.1)]
  5.   2                -1  1      2112  models.common.Conv                      [64, 32, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  6.   3                -2  1      2112  models.common.Conv                      [64, 32, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  7.   4                -1  1      9280  models.common.Conv                      [32, 32, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  8.   5                -1  1      9280  models.common.Conv                      [32, 32, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  9.   6  [-1, -2, -3, -4]  1         0  models.common.Concat                    [1]                           
  10.   7                -1  1      8320  models.common.Conv                      [128, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  11.   8                -1  1         0  models.common.MP                        []                            
  12.   9                -1  1      4224  models.common.Conv                      [64, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  13.  10                -2  1      4224  models.common.Conv                      [64, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  14.  11                -1  1     36992  models.common.Conv                      [64, 64, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  15.  12                -1  1     36992  models.common.Conv                      [64, 64, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  16.  13  [-1, -2, -3, -4]  1         0  models.common.Concat                    [1]                           
  17.  14                -1  1     33024  models.common.Conv                      [256, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  18.  15                -1  1         0  models.common.MP                        []                            
  19.  16                -1  1     16640  models.common.Conv                      [128, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  20.  17                -2  1     16640  models.common.Conv                      [128, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  21.  18                -1  1    147712  models.common.Conv                      [128, 128, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  22.  19                -1  1    147712  models.common.Conv                      [128, 128, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  23.  20  [-1, -2, -3, -4]  1         0  models.common.Concat                    [1]                           
  24.  21                -1  1    131584  models.common.Conv                      [512, 256, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  25.  22                -1  1         0  models.common.MP                        []                            
  26.  23                -1  1     66048  models.common.Conv                      [256, 256, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  27.  24                -2  1     66048  models.common.Conv                      [256, 256, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  28.  25                -1  1    590336  models.common.Conv                      [256, 256, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  29.  26                -1  1    590336  models.common.Conv                      [256, 256, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  30.  27  [-1, -2, -3, -4]  1         0  models.common.Concat                    [1]                           
  31.  28                -1  1    525312  models.common.Conv                      [1024, 512, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  32.  29                -1  1   1086528  models.morepooling.BasicRFB             [512, 256]                    
  33.  30                -1  1     33024  models.common.Conv                      [256, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  34.  31                -1  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
  35.  32                21  1     33024  models.common.Conv                      [256, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  36.  33          [-1, -2]  1         0  models.common.Concat                    [1]                           
  37.  34                -1  1     16512  models.common.Conv                      [256, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  38.  35                -2  1     16512  models.common.Conv                      [256, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  39.  36                -1  1     36992  models.common.Conv                      [64, 64, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  40.  37                -1  1     36992  models.common.Conv                      [64, 64, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  41.  38  [-1, -2, -3, -4]  1         0  models.common.Concat                    [1]                           
  42.  39                -1  1     33024  models.common.Conv                      [256, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  43.  40                -1  1      8320  models.common.Conv                      [128, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  44.  41                -1  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
  45.  42                14  1      8320  models.common.Conv                      [128, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  46.  43          [-1, -2]  1         0  models.common.Concat                    [1]                           
  47.  44                -1  1      4160  models.common.Conv                      [128, 32, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  48.  45                -2  1      4160  models.common.Conv                      [128, 32, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  49.  46                -1  1      9280  models.common.Conv                      [32, 32, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  50.  47                -1  1      9280  models.common.Conv                      [32, 32, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  51.  48  [-1, -2, -3, -4]  1         0  models.common.Concat                    [1]                           
  52.  49                -1  1      8320  models.common.Conv                      [128, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  53.  50                -1  1     73984  models.common.Conv                      [64, 128, 3, 2, None, 1, LeakyReLU(negative_slope=0.1)]
  54.  51          [-1, 39]  1         0  models.common.Concat                    [1]                           
  55.  52                -1  1     16512  models.common.Conv                      [256, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  56.  53                -2  1     16512  models.common.Conv                      [256, 64, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  57.  54                -1  1     36992  models.common.Conv                      [64, 64, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  58.  55                -1  1     36992  models.common.Conv                      [64, 64, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  59.  56  [-1, -2, -3, -4]  1         0  models.common.Concat                    [1]                           
  60.  57                -1  1     33024  models.common.Conv                      [256, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  61.  58                -1  1    295424  models.common.Conv                      [128, 256, 3, 2, None, 1, LeakyReLU(negative_slope=0.1)]
  62.  59          [-1, 29]  1         0  models.common.Concat                    [1]                           
  63.  60                -1  1     65792  models.common.Conv                      [512, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  64.  61                -2  1     65792  models.common.Conv                      [512, 128, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  65.  62                -1  1    147712  models.common.Conv                      [128, 128, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  66.  63                -1  1    147712  models.common.Conv                      [128, 128, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  67.  64  [-1, -2, -3, -4]  1         0  models.common.Concat                    [1]                           
  68.  65                -1  1    131584  models.common.Conv                      [512, 256, 1, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  69.  66                49  1     73984  models.common.Conv                      [64, 128, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  70.  67                57  1    295424  models.common.Conv                      [128, 256, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  71.  68                65  1   1180672  models.common.Conv                      [256, 512, 3, 1, None, 1, LeakyReLU(negative_slope=0.1)]
  72.  69      [66, 67, 68]  1     17132  models.yolo.IDetect                     [1, [[10, 13, 16, 30, 33, 23], [30, 61, 62, 45, 59, 119], [116, 90, 156, 198, 373, 326]], [128, 256, 512]]
  73.  
  74. Model Summary: 284 layers, 6444108 parameters, 6444108 gradients, 13.5 GFLOPS

运行后若打印出如上文本代表改进成功。

四、YOLOv5s改进工作

完成二后，在YOLOv5项目文件下的models文件夹下创建新的文件yolov5s-rfb.yaml，导入如下代码。

  1. # Parameters
  2. nc: 1  # number of classes
  3. depth_multiple: 0.33  # model depth multiple
  4. width_multiple: 0.50  # layer channel multiple
  5. anchors:
  6.   - [10,13, 16,30, 33,23]  # P3/8
  7.   - [30,61, 62,45, 59,119]  # P4/16
  8.   - [116,90, 156,198, 373,326]  # P5/32
  9.  
  10. # YOLOv5 v6.0 backbone
  11. backbone:
  12.   # [from, number, module, args]
  13.   [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
  14.    [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
  15.    [-1, 3, C3, [128]],
  16.    [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
  17.    [-1, 6, C3, [256]],
  18.    [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
  19.    [-1, 9, C3, [512]],
  20.    [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
  21.    [-1, 3, C3, [1024]],
  22.    #[-1, 1, ASPP, [1024]],  # 9
  23.    [-1, 1, BasicRFB, [1024]],
  24.    #[-1, 1, SPP, [1024]],
  25.    #[-1, 1, SPPF, [1024, 5]],  # 9
  26.    #[-1, 1, SPPCSPC, [1024]],
  27.   ]
  28.  
  29. # YOLOv5 v6.0 head
  30. head:
  31.   [[-1, 1, Conv, [512, 1, 1]],
  32.    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
  33.    [[-1, 6], 1, Concat, [1]],  # cat backbone P4
  34.    [-1, 3, C3, [512, False]],  # 13
  35.  
  36.    [-1, 1, Conv, [256, 1, 1]],
  37.    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
  38.    [[-1, 4], 1, Concat, [1]],  # cat backbone P3
  39.    [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
  40.  
  41.    [-1, 1, Conv, [256, 3, 2]],
  42.    [[-1, 14], 1, Concat, [1]],  # cat head P4
  43.    [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
  44.  
  45.    [-1, 1, Conv, [512, 3, 2]],
  46.    [[-1, 10], 1, Concat, [1]],  # cat head P5
  47.    [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
  48.  
  49.    [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
  50.   ]
  1.  
  2.                  from  n    params  module                                  arguments                     
  3.   0                -1  1      3520  models.common.Conv                      [3, 32, 6, 2, 2]              
  4.   1                -1  1     18560  models.common.Conv                      [32, 64, 3, 2]                
  5.   2                -1  1     18816  models.common.C3                        [64, 64, 1]                   
  6.   3                -1  1     73984  models.common.Conv                      [64, 128, 3, 2]               
  7.   4                -1  2    115712  models.common.C3                        [128, 128, 2]                 
  8.   5                -1  1    295424  models.common.Conv                      [128, 256, 3, 2]              
  9.   6                -1  3    625152  models.common.C3                        [256, 256, 3]                 
  10.   7                -1  1   1180672  models.common.Conv                      [256, 512, 3, 2]              
  11.   8                -1  1   1182720  models.common.C3                        [512, 512, 1]                 
  12.   9                -1  1   1316928  models.morepooling.BasicRFB             [512, 512]                    
  13.  10                -1  1    131584  models.common.Conv                      [512, 256, 1, 1]              
  14.  11                -1  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
  15.  12           [-1, 6]  1         0  models.common.Concat                    [1]                           
  16.  13                -1  1    361984  models.common.C3                        [512, 256, 1, False]          
  17.  14                -1  1     33024  models.common.Conv                      [256, 128, 1, 1]              
  18.  15                -1  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
  19.  16           [-1, 4]  1         0  models.common.Concat                    [1]                           
  20.  17                -1  1     90880  models.common.C3                        [256, 128, 1, False]          
  21.  18                -1  1    147712  models.common.Conv                      [128, 128, 3, 2]              
  22.  19          [-1, 14]  1         0  models.common.Concat                    [1]                           
  23.  20                -1  1    296448  models.common.C3                        [256, 256, 1, False]          
  24.  21                -1  1    590336  models.common.Conv                      [256, 256, 3, 2]              
  25.  22          [-1, 10]  1         0  models.common.Concat                    [1]                           
  26.  23                -1  1   1182720  models.common.C3                        [512, 512, 1, False]          
  27.  24      [17, 20, 23]  1     16182  models.yolo.Detect                      [1, [[10, 13, 16, 30, 33, 23], [30, 61, 62, 45, 59, 119], [116, 90, 156, 198, 373, 326]], [128, 256, 512]]
  28.  
  29. Model Summary: 305 layers, 7682358 parameters, 7682358 gradients, 16.5 GFLOPs

运行后若打印出如上文本代表改进成功。

五、YOLOv5n改进工作

完成二后，在YOLOv5项目文件下的models文件夹下创建新的文件yolov5n-rfb.yaml，导入如下代码。

  1. # Parameters
  2. nc: 1  # number of classes
  3. depth_multiple: 0.33  # model depth multiple
  4. width_multiple: 0.25  # layer channel multiple
  5. anchors:
  6.   - [10,13, 16,30, 33,23]  # P3/8
  7.   - [30,61, 62,45, 59,119]  # P4/16
  8.   - [116,90, 156,198, 373,326]  # P5/32
  9.  
  10. # YOLOv5 v6.0 backbone
  11. backbone:
  12.   # [from, number, module, args]
  13.   [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
  14.    [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
  15.    [-1, 3, C3, [128]],
  16.    [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
  17.    [-1, 6, C3, [256]],
  18.    [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
  19.    [-1, 9, C3, [512]],
  20.    [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
  21.    [-1, 3, C3, [1024]],
  22.    #[-1, 1, ASPP, [1024]],  # 9
  23.    [-1, 1, BasicRFB, [1024]],
  24.    #[-1, 1, SPP, [1024]],
  25.    #[-1, 1, SPPF, [1024, 5]],  # 9
  26.    #[-1, 1, SPPCSPC, [1024]],
  27.   ]
  28.  
  29. # YOLOv5 v6.0 head
  30. head:
  31.   [[-1, 1, Conv, [512, 1, 1]],
  32.    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
  33.    [[-1, 6], 1, Concat, [1]],  # cat backbone P4
  34.    [-1, 3, C3, [512, False]],  # 13
  35.  
  36.    [-1, 1, Conv, [256, 1, 1]],
  37.    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
  38.    [[-1, 4], 1, Concat, [1]],  # cat backbone P3
  39.    [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
  40.  
  41.    [-1, 1, Conv, [256, 3, 2]],
  42.    [[-1, 14], 1, Concat, [1]],  # cat head P4
  43.    [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
  44.  
  45.    [-1, 1, Conv, [512, 3, 2]],
  46.    [[-1, 10], 1, Concat, [1]],  # cat head P5
  47.    [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
  48.  
  49.    [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
  50.   ]
  1.  
  2.                  from  n    params  module                                  arguments                     
  3.   0                -1  1      1760  models.common.Conv                      [3, 16, 6, 2, 2]              
  4.   1                -1  1      4672  models.common.Conv                      [16, 32, 3, 2]                
  5.   2                -1  1      4800  models.common.C3                        [32, 32, 1]                   
  6.   3                -1  1     18560  models.common.Conv                      [32, 64, 3, 2]                
  7.   4                -1  2     29184  models.common.C3                        [64, 64, 2]                   
  8.   5                -1  1     73984  models.common.Conv                      [64, 128, 3, 2]               
  9.   6                -1  3    156928  models.common.C3                        [128, 128, 3]                 
  10.   7                -1  1    295424  models.common.Conv                      [128, 256, 3, 2]              
  11.   8                -1  1    296448  models.common.C3                        [256, 256, 1]                 
  12.   9                -1  1    330272  models.morepooling.BasicRFB             [256, 256]                    
  13.  10                -1  1     33024  models.common.Conv                      [256, 128, 1, 1]              
  14.  11                -1  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
  15.  12           [-1, 6]  1         0  models.common.Concat                    [1]                           
  16.  13                -1  1     90880  models.common.C3                        [256, 128, 1, False]          
  17.  14                -1  1      8320  models.common.Conv                      [128, 64, 1, 1]               
  18.  15                -1  1         0  torch.nn.modules.upsampling.Upsample    [None, 2, 'nearest']          
  19.  16           [-1, 4]  1         0  models.common.Concat                    [1]                           
  20.  17                -1  1     22912  models.common.C3                        [128, 64, 1, False]           
  21.  18                -1  1     36992  models.common.Conv                      [64, 64, 3, 2]                
  22.  19          [-1, 14]  1         0  models.common.Concat                    [1]                           
  23.  20                -1  1     74496  models.common.C3                        [128, 128, 1, False]          
  24.  21                -1  1    147712  models.common.Conv                      [128, 128, 3, 2]              
  25.  22          [-1, 10]  1         0  models.common.Concat                    [1]                           
  26.  23                -1  1    296448  models.common.C3                        [256, 256, 1, False]          
  27.  24      [17, 20, 23]  1      8118  models.yolo.Detect                      [1, [[10, 13, 16, 30, 33, 23], [30, 61, 62, 45, 59, 119], [116, 90, 156, 198, 373, 326]], [64, 128, 256]]
  28.  
  29. Model Summary: 305 layers, 1930934 parameters, 1930934 gradients, 4.4 GFLOPs
六、代码部分的解释

BasicRFB 类继承自 PyTorch 的 nn.Module。构造函数接收输入通道数 in_planes、输出通道数 out_planes 和一些控制模块行为的参数，比如步长 stride、缩放因子 scale、中间层通道减少倍数 map_reduce、扩张率 vision 和分组卷积的组数 groups。

三个分支分别定义为 branch0、branch1 和 branch2。每个分支都由多个 BasicConv 组件构成，其中包含卷积、批量归一化和激活函数。分支的内核大小、填充和扩张率各不相同，目的是捕捉不同尺度的信息。branch0 使用标准卷积和扩张卷积；branch1 使用稍大的扩张率；而 branch2 则使用更大的扩张率和额外的卷积层，进一步增加感受野的多样性。

最终，三个分支的输出被拼接在一起，并通过 ConvLinear 卷积层转换为期望的输出通道数。此外，还有一个快捷连接（shortcut）从输入直接传递到输出，与 ResNet 结构类似，用于缓解梯度消失问题。所有分支的输出和快捷连接的输出在相加前会乘以一个缩放因子 scale，然后经过 ReLU 激活函数，得到最终的输出。

整个 BasicRFB 模块在前向传播时，将输入通过三个分支并行处理，然后将结果拼接后进行线性变换，最后加上快捷连接的输出。这种设计使得模块能够高效地捕获多尺度特征，同时保持实时处理速度，适用于目标检测等计算机视觉任务。

运行后打印如上代码说明改进成功。

更多文章产出中，主打简洁和准确，欢迎关注我，共同探讨！

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/Guff_9hys/article/detail/802180
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号