---

---

👀🎉📜系列文章目录

必读【论文精读】DAMO-YOLO：兼顾速度与精度的高效目标检测框架（ A Report on Real-Time Object Detection Design）
【YOLOv5改进系列(1)】高效涨点----使用EIoU、Alpha-IoU、SIoU、Focal-EIOU替换CIou
【YOLOv5改进系列(2)】高效涨点----Wise-IoU详细解读及使用Wise-IoU(WIOU)替换CIOU
【YOLOv5改进系列(3)】高效涨点----Optimal Transport Assignment：OTA最优传输方法
【YOLOv5改进系列(4)】高效涨点----添加可变形卷积DCNv2
【YOLOv5改进系列(5)】高效涨点----添加密集小目标检测NWD方法

🚀🚀🚀前言

⚡️DAMO-YOLO是阿里巴巴达摩院在2022年提出的一种模型，其实在当时yolov5和v6已经出来了，并且也有部分实时监测比较好的算法模型，但是当前检测框架在实际应用时仍然有以下几个痛点：

• ① 模型尺度变化不够灵活，难以适应不同的算力场景。如 YOLO 系列的检测框架，一般只提供 3-5 个模型的计算量，从十几到一百多 Flops 数量级，难以覆盖不同的算力场景。
• ② 多尺度检测能力弱，特别是小物体检测性能较差，这使得模型应用场景十分受限。比如在无人机检测场景，它们的效果往往都不太理想。
• ③ 速度/精度曲线不够理想，速度和精度难以同时兼容。

☀️针对上述情况，达摩院计算机视觉团队设计并开源了 DAMO-YOLO，DAMO-YOLO 主要着眼于工业落地。相比于其他的目标检测框架具有三个明显的技术优势：

• ① 整合了自研 NAS 技术，可低成本自定义模型，让用户充分发挥芯片算力。
• ② 结合 Efficient RepGFPN 以及 HeavyNeck 模型设计范式，能够很大程度上提高模型的多尺度检测能力，扩大模型应用范围。
• ③ 提出了全尺度通用的蒸馏技术，将大模型的知识转移到小模型上，在不带来推理负担的情况下，提升小模型的性能。

关于DAMO-YOLO的一些其他细节这里不做过多解释，像MAE-NAS技术进行 backbone 的搜索、蒸馏模型等技术如果感兴趣可以看一下这篇文章DAMO-YOLO：【论文精读】DAMO-YOLO：兼顾速度与精度的高效目标检测框架（ A Report on Real-Time Object Detection Design）

🚀本文改进使用的是DAMO-YOLO 中的Efficient RepGFPN特征网络融合部分，该网络是在GFPN特征网络上进行改进的，因为GFPN有很多上采样操作并且并行化比较低，导致了Flops高效Latency低效；因此我使用Efficient RepGFPN方法去替换了yolov5中Neck部分的PANet融合方法；

实验方面我分别使用yolov5s.pt和yolov5l.pt两个权重来进行训练，然后将yolov5_GFPN.yaml的depth_multiple和width_multiple分别设置成（0.33，0.50）和（1.0,1.0）对标yolov5s和yolov5l。
实验结果： 与yolov5s基准模型相比，使用Efficient RepGFPN之后map@0.5反而下降了，但是替换之后与yolov5l基准模型相比，map@0.5提升了近4个百分点，同时f1置信分数也有所增加。至于为什么会出现这种情况，在实验总结部分我写出自己理解的一些解释。

---

一、1️⃣ 添加yolov5_GFPN.yaml文件

🚀在models文件下新建一个yolov5_GFPN.yaml文件，将如下代码复制到yaml文件中；我这里depth_multiple、width_multiple设置的都是1，需要使用yolov5l权重进行训练，如果想要使用yolov5s的权重训练，只需要将depth_multiple、width_multiple分别设置为0.33和0.5。

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license# Parameters
nc: 6  # number of classes
depth_multiple: 1.00  # model depth multiple
width_multiple: 1.00  # 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 v6.0 backbone
backbone:# [from, number, module, args][ [ -1, 1, Conv, [ 64, 6, 2, 2 ] ],  # 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, 6, 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, 3, C3, [ 1024 ] ],[ -1, 1, SPPF, [ 1024, 5 ] ],  # 9]# DAMO-YOLO GFPN Head
head:[ [ -1, 1, Conv, [ 512, 1, 1 ] ], # 10 添加这个是为了使用1x1卷积进行降维[ 6, 1, Conv, [ 512, 3, 2 ] ],[ [ -1, 10 ], 1, Concat, [ 1 ] ],[ -1, 3, CSPStage, [ 512 ] ], # 13[ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], #14[ 4, 1, Conv, [ 256, 3, 2 ] ], # 15[ [ 14, -1, 6 ], 1, Concat, [ 1 ] ],[ -1, 3, CSPStage, [ 512 ] ], # 17[ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],[ [ -1, 4 ], 1, Concat, [ 1 ] ],[ -1, 3, CSPStage, [ 256 ] ], # 20[ -1, 1, Conv, [ 256, 3, 2 ] ],[ [ -1, 17 ], 1, Concat, [ 1 ] ],[ -1, 3, CSPStage, [ 512 ] ], # 23[ 17, 1, Conv, [ 256, 3, 2 ] ], # 24[ 23, 1, Conv, [ 256, 3, 2 ] ], # 25[ [ 13, 24, -1 ], 1, Concat, [ 1 ] ],[ -1, 3, CSPStage, [ 1024 ] ], # 27[ [ 20, 23, 27 ], 1, Detect, [ nc, anchors ] ],  # Detect(P3, P4, P5)]

二、2️⃣添加extra_modules.py代码

📌在models文件下新建一个extra_modules.py文件，将如下代码复制到文件中；这部分代码就是Efficient RepGFPN的实现方法。

import torch
import torch.nn as nn
import torch.nn.functional as Fdef conv_bn(in_channels, out_channels, kernel_size, stride, padding, groups=1):'''Basic cell for rep-style block, including conv and bn'''result = nn.Sequential()result.add_module('conv',nn.Conv2d(in_channels=in_channels,out_channels=out_channels,kernel_size=kernel_size,stride=stride,padding=padding,groups=groups,bias=False))result.add_module('bn', nn.BatchNorm2d(num_features=out_channels))return resultclass RepConv(nn.Module):'''RepConv is a basic rep-style block, including training and deploy statusCode is based on https://github.com/DingXiaoH/RepVGG/blob/main/repvgg.py'''def __init__(self,in_channels,out_channels,kernel_size=3,stride=1,padding=1,dilation=1,groups=1,padding_mode='zeros',deploy=False,act='relu',norm=None):super(RepConv, self).__init__()self.deploy = deployself.groups = groupsself.in_channels = in_channelsself.out_channels = out_channelsassert kernel_size == 3assert padding == 1padding_11 = padding - kernel_size // 2if isinstance(act, str):self.nonlinearity = get_activation(act)else:self.nonlinearity = actif deploy:self.rbr_reparam = nn.Conv2d(in_channels=in_channels,out_channels=out_channels,kernel_size=kernel_size,stride=stride,padding=padding,dilation=dilation,groups=groups,bias=True,padding_mode=padding_mode)else:self.rbr_identity = Noneself.rbr_dense = conv_bn(in_channels=in_channels,out_channels=out_channels,kernel_size=kernel_size,stride=stride,padding=padding,groups=groups)self.rbr_1x1 = conv_bn(in_channels=in_channels,out_channels=out_channels,kernel_size=1,stride=stride,padding=padding_11,groups=groups)def forward(self, inputs):'''Forward process'''if hasattr(self, 'rbr_reparam'):return self.nonlinearity(self.rbr_reparam(inputs))if self.rbr_identity is None:id_out = 0else:id_out = self.rbr_identity(inputs)return self.nonlinearity(self.rbr_dense(inputs) + self.rbr_1x1(inputs) + id_out)def get_equivalent_kernel_bias(self):kernel3x3, bias3x3 = self._fuse_bn_tensor(self.rbr_dense)kernel1x1, bias1x1 = self._fuse_bn_tensor(self.rbr_1x1)kernelid, biasid = self._fuse_bn_tensor(self.rbr_identity)return kernel3x3 + self._pad_1x1_to_3x3_tensor(kernel1x1) + kernelid, bias3x3 + bias1x1 + biasiddef _pad_1x1_to_3x3_tensor(self, kernel1x1):if kernel1x1 is None:return 0else:return torch.nn.functional.pad(kernel1x1, [1, 1, 1, 1])def _fuse_bn_tensor(self, branch):if branch is None:return 0, 0if isinstance(branch, nn.Sequential):kernel = branch.conv.weightrunning_mean = branch.bn.running_meanrunning_var = branch.bn.running_vargamma = branch.bn.weightbeta = branch.bn.biaseps = branch.bn.epselse:assert isinstance(branch, nn.BatchNorm2d)if not hasattr(self, 'id_tensor'):input_dim = self.in_channels // self.groupskernel_value = np.zeros((self.in_channels, input_dim, 3, 3),dtype=np.float32)for i in range(self.in_channels):kernel_value[i, i % input_dim, 1, 1] = 1self.id_tensor = torch.from_numpy(kernel_value).to(branch.weight.device)kernel = self.id_tensorrunning_mean = branch.running_meanrunning_var = branch.running_vargamma = branch.weightbeta = branch.biaseps = branch.epsstd = (running_var + eps).sqrt()t = (gamma / std).reshape(-1, 1, 1, 1)return kernel * t, beta - running_mean * gamma / stddef switch_to_deploy(self):if hasattr(self, 'rbr_reparam'):returnkernel, bias = self.get_equivalent_kernel_bias()self.rbr_reparam = nn.Conv2d(in_channels=self.rbr_dense.conv.in_channels,out_channels=self.rbr_dense.conv.out_channels,kernel_size=self.rbr_dense.conv.kernel_size,stride=self.rbr_dense.conv.stride,padding=self.rbr_dense.conv.padding,dilation=self.rbr_dense.conv.dilation,groups=self.rbr_dense.conv.groups,bias=True)self.rbr_reparam.weight.data = kernelself.rbr_reparam.bias.data = biasfor para in self.parameters():para.detach_()self.__delattr__('rbr_dense')self.__delattr__('rbr_1x1')if hasattr(self, 'rbr_identity'):self.__delattr__('rbr_identity')if hasattr(self, 'id_tensor'):self.__delattr__('id_tensor')self.deploy = Trueclass Swish(nn.Module):def __init__(self, inplace=True):super(Swish, self).__init__()self.inplace = inplacedef forward(self, x):if self.inplace:x.mul_(F.sigmoid(x))return xelse:return x * F.sigmoid(x)def get_activation(name='silu', inplace=True):if name is None:return nn.Identity()if isinstance(name, str):if name == 'silu':module = nn.SiLU(inplace=inplace)elif name == 'relu':module = nn.ReLU(inplace=inplace)elif name == 'lrelu':module = nn.LeakyReLU(0.1, inplace=inplace)elif name == 'swish':module = Swish(inplace=inplace)elif name == 'hardsigmoid':module = nn.Hardsigmoid(inplace=inplace)elif name == 'identity':module = nn.Identity()else:raise AttributeError('Unsupported act type: {}'.format(name))return moduleelif isinstance(name, nn.Module):return nameelse:raise AttributeError('Unsupported act type: {}'.format(name))def get_norm(name, out_channels, inplace=True):if name == 'bn':module = nn.BatchNorm2d(out_channels)else:raise NotImplementedErrorreturn moduleclass ConvBNAct(nn.Module):"""A Conv2d -> Batchnorm -> silu/leaky relu block"""def __init__(self,in_channels,out_channels,ksize,stride=1,groups=1,bias=False,act='silu',norm='bn',reparam=False,):super().__init__()# same paddingpad = (ksize - 1) // 2self.conv = nn.Conv2d(in_channels,out_channels,kernel_size=ksize,stride=stride,padding=pad,groups=groups,bias=bias,)if norm is not None:self.bn = get_norm(norm, out_channels, inplace=True)if act is not None:self.act = get_activation(act, inplace=True)self.with_norm = norm is not Noneself.with_act = act is not Nonedef forward(self, x):x = self.conv(x)if self.with_norm:x = self.bn(x)if self.with_act:x = self.act(x)return xdef fuseforward(self, x):return self.act(self.conv(x))class BasicBlock_3x3_Reverse(nn.Module):def __init__(self,ch_in,ch_hidden_ratio,ch_out,act='relu',shortcut=True):super(BasicBlock_3x3_Reverse, self).__init__()assert ch_in == ch_outch_hidden = int(ch_in * ch_hidden_ratio)self.conv1 = ConvBNAct(ch_hidden, ch_out, 3, stride=1, act=act)self.conv2 = RepConv(ch_in, ch_hidden, 3, stride=1, act=act)self.shortcut = shortcutdef forward(self, x):y = self.conv2(x)y = self.conv1(y)if self.shortcut:return x + yelse:return yclass SPP(nn.Module):def __init__(self,ch_in,ch_out,k,pool_size,act='swish',):super(SPP, self).__init__()self.pool = []for i, size in enumerate(pool_size):pool = nn.MaxPool2d(kernel_size=size,stride=1,padding=size // 2,ceil_mode=False)self.add_module('pool{}'.format(i), pool)self.pool.append(pool)self.conv = ConvBNAct(ch_in, ch_out, k, act=act)def forward(self, x):outs = [x]for pool in self.pool:outs.append(pool(x))y = torch.cat(outs, axis=1)y = self.conv(y)return yclass CSPStage(nn.Module):def __init__(self,ch_in,ch_out,n,block_fn='BasicBlock_3x3_Reverse',ch_hidden_ratio=1.0,act='silu',spp=False):super(CSPStage, self).__init__()split_ratio = 2ch_first = int(ch_out // split_ratio)ch_mid = int(ch_out - ch_first)self.conv1 = ConvBNAct(ch_in, ch_first, 1, act=act)self.conv2 = ConvBNAct(ch_in, ch_mid, 1, act=act)self.convs = nn.Sequential()next_ch_in = ch_midfor i in range(n):if block_fn == 'BasicBlock_3x3_Reverse':self.convs.add_module(str(i),BasicBlock_3x3_Reverse(next_ch_in,ch_hidden_ratio,ch_mid,act=act,shortcut=True))else:raise NotImplementedErrorif i == (n - 1) // 2 and spp:self.convs.add_module('spp', SPP(ch_mid * 4, ch_mid, 1, [5, 9, 13], act=act))next_ch_in = ch_midself.conv3 = ConvBNAct(ch_mid * n + ch_first, ch_out, 1, act=act)def forward(self, x):y1 = self.conv1(x)y2 = self.conv2(x)mid_out = [y1]for conv in self.convs:y2 = conv(y2)mid_out.append(y2)y = torch.cat(mid_out, axis=1)y = self.conv3(y)return y

三、3️⃣yolo.py文件添加内容

3.1 🎓 添加CSPStage模块

📌找到models文件夹的yolo.py文件，在最上方将extra_modules.py文件中的CSPStage模块导入进来，代码如下：

from models.extra_modules import CSPStage

📌然后在yolo.py文件中找到parse_model网络解析函数，在下面两个地方添加CSPStage；

四、4️⃣实验结果

4.1 🎓 使用yolov5s.pt训练的结果对比

yolov5基准模型训练结果：F1置信度分数为0.71、map@0.5=0.78；

添加Efficient RepGFPN模块训练结果：F1置信度分数为0.76、map@0.5=0.763；

4.2 ✨ 使用yolov5l.pt训练的结果对比

yolov5基准模型训练结果：F1置信度分数为0.8、map@0.5=0.795；

添加Efficient RepGFPN模块训练结果：F1置信度分数为0.82、map@0.5=0.833；

4.3 ⭐️ 实验结果分析

🚀两个对比实验可以看出，使用yolov5s.pt进行训练的时候，Efficient RepGFPN模块替换原有的PANet之后map@0.5反而下降了，但是yolov5l.pt相较于基准模型，map@0.5和f1置信分数都有明显的增加，这个可能是和Efficient RepGFPN的HeavyNeck有关，DAMO-YOLO设计的理念是将小部分参数运算应用到backbone部分，将大部分的参数运算放在了Neck特征融合部分，所以在网络加深的情况下，Neck可以获取到更多的特征信息，所以在处理特征信息方面可能要优于PANet。欢迎大家一起讨论！！！

---