一、本文介绍

本文记录的是 利用 CGNet 中的 CG block 模块优化 RT-DETR 的目标检测网络模型 。 CG block 通过 局部特征提取器 、 周围环境提取器 、 联合特征提取器 和 全局环境提取器 来提取 局部特征 、 周围环境 和 全局环境信息 ， 充分利用不同层次的信息 。本文将其应用到 RT-DETR 中，并进行 二次创新 ，使网络能够更好的 处理多类别目标的分类能力。

---

---

二、CGblock 介绍

CGNet: A Light-Weight Context Guided Network for Semantic Segmentation

2.1 设计出发点

• 受到人类视觉系统依赖上下文信息理解场景的启发。例如人类视觉系统识别黄色区域时，如果只关注该区域本身很难识别，但结合其周围环境（红色区域）以及整个场景的全局环境（紫色区域）就更容易对黄色区域进行分类，
• 所以设计 CG block 来充分利用局部特征、周围环境和全局环境以提高准确性。

2.2 原理

1️⃣ 首先通过 局部特征提取器 $floc(\ast )$ 和周围环境提取器 $f_{sur}(\ast )$ 分别学习 局部特征 和 周围环境特征 ；

2️⃣ 然后由 联合特征提取器 $f_{joi}(\ast )$ 获得 联合特征 ；

3️⃣ 最后利用 全局环境提取器 $folo(\ast )$ 提取 全局环境来改进联合特征 。

2.3 结构

• 如 图(d) 所示， CG block 由以下部分组成：
  • 局部特征提取器 $floc(\ast )$ : 实例化为 3×3标准卷积层 ，从8个相邻特征向量中 学习局部特征 ，对应 图(a) 中的黄色区域。
  • 周围环境提取器 $f_{sur}(\ast )$ : 实例化为 3×3空洞/扩张卷积层 ，因为空洞/扩张卷积有较大的感受野能有效 学习周围环境 ，对应 图(b) 中的红色区域。
  • 联合特征提取器 $f_{joi}(\ast )$ : 设计为一个 连接层 ，后面跟着 批量归一化（BN） 和 参数化ReLU（PReLU） 操作符，从 $floc(\ast )$ 和 $f_{sur}(\ast )$ 的输出中 获取联合特征 。
  • 全局环境提取器 $folo(\ast )$ : 实例化为一个 全局平均池化层 来聚合对应 图(c) 中紫色区域的 全局环境 ，后面跟着一个多层感知器进一步提取全局环境，最后通过一个尺度层用提取的全局环境对联合特征进行重新加权。

• 此外， CG block 还采用了 两种残差连接 ：
  • 局部残差学习（LRL） ：连接输入和联合特征提取器 $f_{joi}(\ast )$ 。
  • 全局残差学习（GRL） ：连接输入和全局特征提取器 $folo(\ast )$ ， GRL比LRL在促进网络信息流动方面能力更强。

2.4 优势

• 有效利用多种信息 ：充分利用了 局部特征 、 周围环境 和 全局环境 ，能更好地对物体进行分类，提高分割准确性。
• 残差学习促进训练 ：采用 残差学习 有助于 学习高度复杂的特征 ，并在训练过程中 改善梯度反向传播 。
• 合理的结构设计 ：各部分的实例化选择合理，如周围环境提取器采用空洞/扩张卷积层，全局环境提取器采用全局平均池化层和多层感知器的组合等。

论文： https://arxiv.org/pdf/1811.08201.pdf
源码： https://github.com/wutianyiRosun/CGNet

三、CGblock的实现代码

CGblock 及其改进的实现代码如下：

import torch
import torch.nn as nn
import torch.nn.functional as F

from ultralytics.nn.modules.conv import LightConv

# https://arxiv.org/pdf/1811.08201.pdf
# https://github.com/wutianyiRosun/CGNet
 
class ConvBNPReLU(nn.Module):
    def __init__(self, nIn, nOut, kSize, stride=1):
        """
        args:
            nIn: number of input channels
            nOut: number of output channels
            kSize: kernel size
            stride: stride rate for down-sampling. Default is 1
        """
        super().__init__()
        if isinstance(kSize, tuple):
            kSize = kSize[0]
        padding = int((kSize - 1) / 2)
        self.conv = nn.Conv2d(nIn, nOut, (kSize, kSize), stride=stride, padding=(padding, padding), bias=False)
        self.bn = nn.BatchNorm2d(nOut, eps=1e-03)
        self.act = nn.PReLU(nOut)
 
    def forward(self, input):
        """
        args:
           input: input feature map
           return: transformed feature map
        """
        output = self.conv(input)
        output = self.bn(output)
        output = self.act(output)
        return output

class BNPReLU(nn.Module):
    def __init__(self, nOut):
        """
        args:
           nOut: channels of output feature maps
        """
        super().__init__()
        self.bn = nn.BatchNorm2d(nOut, eps=1e-03)
        self.act = nn.PReLU(nOut)
 
    def forward(self, input):
        """
        args:
           input: input feature map
           return: normalized and thresholded feature map
        """
        output = self.bn(input)
        output = self.act(output)
        return output
 
class ConvBN(nn.Module):
    def __init__(self, nIn, nOut, kSize, stride=1):
        """
        args:
           nIn: number of input channels
           nOut: number of output channels
           kSize: kernel size
           stride: optinal stide for down-sampling
        """
        super().__init__()
        if isinstance(kSize, tuple):
            kSize = kSize[0]
        padding = int((kSize - 1) / 2)
        self.conv = nn.Conv2d(nIn, nOut, (kSize, kSize), stride=stride, padding=(padding, padding), bias=False)
        self.bn = nn.BatchNorm2d(nOut, eps=1e-03)
 
    def forward(self, input):
        """
        args:
           input: input feature map
           return: transformed feature map
        """
        output = self.conv(input)
        output = self.bn(output)
        return output

class Conv(nn.Module):
    def __init__(self, nIn, nOut, kSize, stride=1):
        """
        args:
            nIn: number of input channels
            nOut: number of output channels
            kSize: kernel size
            stride: optional stride rate for down-sampling
        """
        super().__init__()
        if isinstance(kSize, tuple):
            kSize = kSize[0]
        padding = int((kSize - 1) / 2)
        self.conv = nn.Conv2d(nIn, nOut, (kSize, kSize), stride=stride, padding=(padding, padding), bias=False)
 
    def forward(self, input):
        """
        args:
           input: input feature map
           return: transformed feature map
        """
        output = self.conv(input)
        return output
 
class ChannelWiseConv(nn.Module):
    def __init__(self, nIn, nOut, kSize, stride=1):
        """
        Args:
            nIn: number of input channels
            nOut: number of output channels, default (nIn == nOut)
            kSize: kernel size
            stride: optional stride rate for down-sampling
        """
        super().__init__()
        if isinstance(kSize, tuple):
            kSize = kSize[0]
        padding = int((kSize - 1) / 2)
        self.conv = nn.Conv2d(nIn, nOut, (kSize, kSize), stride=stride, padding=(padding, padding), groups=nIn,
                              bias=False)
 
    def forward(self, input):
        """
        args:
           input: input feature map
           return: transformed feature map
        """
        output = self.conv(input)
        return output

class DilatedConv(nn.Module):
    def __init__(self, nIn, nOut, kSize, stride=1, d=1):
        """
        args:
           nIn: number of input channels
           nOut: number of output channels
           kSize: kernel size
           stride: optional stride rate for down-sampling
           d: dilation rate
        """
        super().__init__()
        if isinstance(kSize, tuple):
            kSize = kSize[0]
        padding = int((kSize - 1) / 2) * d
        self.conv = nn.Conv2d(nIn, nOut, (kSize, kSize), stride=stride, padding=(padding, padding), bias=False,
                              dilation=d)
 
    def forward(self, input):
        """
        args:
           input: input feature map
           return: transformed feature map
        """
        output = self.conv(input)
        return output

class ChannelWiseDilatedConv(nn.Module):
    def __init__(self, nIn, nOut, kSize, stride=1, d=1):
        """
        args:
           nIn: number of input channels
           nOut: number of output channels, default (nIn == nOut)
           kSize: kernel size
           stride: optional stride rate for down-sampling
           d: dilation rate
        """
        super().__init__()
        if isinstance(kSize, tuple):
            kSize = kSize[0]
        padding = int((kSize - 1) / 2) * d
        self.conv = nn.Conv2d(nIn, nOut, (kSize, kSize), stride=stride, padding=(padding, padding), groups=nIn,
                              bias=False, dilation=d)
 
    def forward(self, input):
        """
        args:
           input: input feature map
           return: transformed feature map
        """
        output = self.conv(input)
        return output
 
class FGlo(nn.Module):
    """
    the FGlo class is employed to refine the joint feature of both local feature and surrounding context.
    """
 
    def __init__(self, channel, reduction=16):
        super(FGlo, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
            nn.Linear(channel, channel // reduction),
            nn.ReLU(inplace=True),
            nn.Linear(channel // reduction, channel),
            nn.Sigmoid()
        )
 
    def forward(self, x):
        b, c, _, _ = x.size()
        y = self.avg_pool(x).view(b, c)
        y = self.fc(y).view(b, c, 1, 1)
        return x * y
 
class ContextGuidedBlock_Down(nn.Module):
    """
    the size of feature map divided 2, (H,W,C)---->(H/2, W/2, 2C)
    """
 
    def __init__(self, nIn, nOut, dilation_rate=2, reduction=16):
        """
        args:
           nIn: the channel of input feature map
           nOut: the channel of output feature map, and nOut=2*nIn
        """
        super().__init__()
 
        self.conv1x1 = ConvBNPReLU(nIn, nOut, 3, 2)  # size/2, channel: nIn--->nOut
 
        self.F_loc = ChannelWiseConv(nOut, nOut, 3, 1)
        self.F_sur = ChannelWiseDilatedConv(nOut, nOut, 3, 1, dilation_rate)
 
        self.bn = nn.BatchNorm2d(2 * nOut, eps=1e-3)
        self.act = nn.PReLU(2 * nOut)
        self.reduce = Conv(2 * nOut, nOut, 1, 1)  # reduce dimension: 2*nOut--->nOut
 
        self.F_glo = FGlo(nOut, reduction)
 
    def forward(self, input):
        output = self.conv1x1(input)
        loc = self.F_loc(output)
        sur = self.F_sur(output)
 
        joi_feat = torch.cat([loc, sur], 1)  # the joint feature
        joi_feat = self.bn(joi_feat)
        joi_feat = self.act(joi_feat)
        joi_feat = self.reduce(joi_feat)  # channel= nOut
 
        output = self.F_glo(joi_feat)  # F_glo is employed to refine the joint feature
 
        return output
 
class ContextGuidedBlock(nn.Module):
    def __init__(self, nIn, nOut, dilation_rate=2, reduction=16, add=True):
        """
        args:
           nIn: number of input channels
           nOut: number of output channels,
           add: if true, residual learning
        """
        super().__init__()
        n = int(nOut / 2)
        self.conv1x1 = ConvBNPReLU(nIn, n, 1, 1)  # 1x1 Conv is employed to reduce the computation
        self.F_loc = ChannelWiseConv(n, n, 3, 1)  # local feature
        self.F_sur = ChannelWiseDilatedConv(n, n, 3, 1, dilation_rate)  # surrounding context
        self.bn_prelu = BNPReLU(nOut)
        self.add = add
        self.F_glo = FGlo(nOut, reduction)
 
    def forward(self, input):
        output = self.conv1x1(input)
        loc = self.F_loc(output)
        sur = self.F_sur(output)
 
        joi_feat = torch.cat([loc, sur], 1)
 
        joi_feat = self.bn_prelu(joi_feat)
 
        output = self.F_glo(joi_feat)  # F_glo is employed to refine the joint feature
        # if residual version
        if self.add:
            output = input + output
        return output
 
def autopad(k, p=None, d=1):  # kernel, padding, dilation
    """Pad to 'same' shape outputs."""
    if d > 1:
        k = d * (k - 1) + 1 if isinstance(k, int) else [d * (x - 1) + 1 for x in k]  # actual kernel-size
    if p is None:
        p = k // 2 if isinstance(k, int) else [x // 2 for x in k]  # auto-pad
    return p

class Conv(nn.Module):
    """Standard convolution with args(ch_in, ch_out, kernel, stride, padding, groups, dilation, activation)."""
 
    default_act = nn.SiLU()  # default activation
 
    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, d=1, act=True):
        """Initialize Conv layer with given arguments including activation."""
        super().__init__()
        self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p, d), groups=g, dilation=d, bias=False)
        self.bn = nn.BatchNorm2d(c2)
        self.act = self.default_act if act is True else act if isinstance(act, nn.Module) else nn.Identity()
 
    def forward(self, x):
        """Apply convolution, batch normalization and activation to input tensor."""
        return self.act(self.bn(self.conv(x)))
 
    def forward_fuse(self, x):
        """Perform transposed convolution of 2D data."""
        return self.act(self.conv(x))

class HGBlock_ContextGuidedBlock(nn.Module):
    """
    HG_Block of PPHGNetV2 with 2 convolutions and LightConv.

    https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/backbones/hgnet_v2.py
    """

    def __init__(self, c1, cm, c2, k=3, n=6, lightconv=False, shortcut=False, act=nn.ReLU()):
        """Initializes a CSP Bottleneck with 1 convolution using specified input and output channels."""
        super().__init__()
        block = LightConv if lightconv else Conv
        self.m = nn.ModuleList(block(c1 if i == 0 else cm, cm, k=k, act=act) for i in range(n))
        self.sc = Conv(c1 + n * cm, c2 // 2, 1, 1, act=act)  # squeeze conv
        self.ec = Conv(c2 // 2, c2, 1, 1, act=act)  # excitation conv
        self.add = shortcut and c1 == c2
        self.cv = ContextGuidedBlock(c2, c2)
        
    def forward(self, x):
        """Forward pass of a PPHGNetV2 backbone layer."""
        y = [x]
        y.extend(m(y[-1]) for m in self.m)
        y = self.cv(self.ec(self.sc(torch.cat(y, 1))))
        return y + x if self.add else y

---

四、创新模块

4.1 改进点1⭐

模块改进方法 ：直接加入 ContextGuidedBlock_Down模块 （ 第五节讲解添加步骤 ）。

ContextGuidedBlock_Down模块 添加后如下：

4.2 改进点2⭐

模块改进方法 ：基于 ContextGuidedBlock模块 的 HGBlock （ 第五节讲解添加步骤 ）。

第二种改进方法是对 RT-DETR 中的 HGBlock模块 进行改进，并将 ContextGuidedBlock 在加入到 HGBlock 模块中。

改进代码如下：

首先，在 HGBlock 中加入 ContextGuidedBlock模块 ，并重命名为 HGBlock_ContextGuidedBlock 。

class HGBlock_ContextGuidedBlock(nn.Module):
    """
    HG_Block of PPHGNetV2 with 2 convolutions and LightConv.

    https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/backbones/hgnet_v2.py
    """

    def __init__(self, c1, cm, c2, k=3, n=6, lightconv=False, shortcut=False, act=nn.ReLU()):
        """Initializes a CSP Bottleneck with 1 convolution using specified input and output channels."""
        super().__init__()
        block = LightConv if lightconv else Conv
        self.m = nn.ModuleList(block(c1 if i == 0 else cm, cm, k=k, act=act) for i in range(n))
        self.sc = Conv(c1 + n * cm, c2 // 2, 1, 1, act=act)  # squeeze conv
        self.ec = Conv(c2 // 2, c2, 1, 1, act=act)  # excitation conv
        self.add = shortcut and c1 == c2
        self.cv = ContextGuidedBlock(c2, c2)
        
    def forward(self, x):
        """Forward pass of a PPHGNetV2 backbone layer."""
        y = [x]
        y.extend(m(y[-1]) for m in self.m)
        y = self.cv(self.ec(self.sc(torch.cat(y, 1))))
        return y + x if self.add else y

注意❗：在 第五小节 中需要声明的模块名称为： ContextGuidedBlock_Down 和 HGBlock_ContextGuidedBlock 。

---

五、添加步骤

5.1 修改一

① 在 ultralytics/nn/ 目录下新建 AddModules 文件夹用于存放模块代码

② 在 AddModules 文件夹下新建 CGNetBlock.py ，将 第三节 中的代码粘贴到此处

5.2 修改二

在 AddModules 文件夹下新建 __init__.py （已有则不用新建），在文件内导入模块： from .CGNetBlock import *

5.3 修改三

在 ultralytics/nn/modules/tasks.py 文件中，需要在两处位置添加各模块类名称。

首先：导入模块

其次：在 parse_model函数 中注册 ContextGuidedBlock_Down 和 HGBlock_ContextGuidedBlock 模块

---

六、yaml模型文件

6.1 模型改进版本1

此处以 ultralytics/cfg/models/rt-detr/rtdetr-l.yaml 为例，在同目录下创建一个用于自己数据集训练的模型文件 rtdetr-l-ContextGuidedBlock_Down.yaml 。

将 rtdetr-l.yaml 中的内容复制到 rtdetr-l-ContextGuidedBlock_Down.yaml 文件下，修改 nc 数量等于自己数据中目标的数量。

📌 模型的修改方法是将骨干网络中的 下采样模块 替换成 ContextGuidedBlock_Down模块 。

# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-l object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/models/rtdetr

# Parameters
nc: 1 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
  # [depth, width, max_channels]
  l: [1.00, 1.00, 1024]

backbone:
  # [from, repeats, module, args]
  - [-1, 1, HGStem, [32, 48]] # 0-P2/4
  - [-1, 6, HGBlock, [48, 128, 3]] # stage 1

  - [-1, 1, ContextGuidedBlock_Down, [128]] # 2-P3/8
  - [-1, 6, HGBlock, [96, 512, 3]] # stage 2

  - [-1, 1, ContextGuidedBlock_Down, [512]] # 4-P4/16
  - [-1, 6, HGBlock, [192, 1024, 5, True, False]] # cm, c2, k, light, shortcut
  - [-1, 6, HGBlock, [192, 1024, 5, True, True]]
  - [-1, 6, HGBlock, [192, 1024, 5, True, True]] # stage 3

  - [-1, 1, ContextGuidedBlock_Down, [1024]] # 8-P5/32
  - [-1, 6, HGBlock, [384, 2048, 5, True, False]] # stage 4

head:
  - [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 10 input_proj.2
  - [-1, 1, AIFI, [1024, 8]]
  - [-1, 1, Conv, [256, 1, 1]] # 12, Y5, lateral_convs.0

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
  - [7, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 14 input_proj.1
  - [[-2, -1], 1, Concat, [1]]
  - [-1, 3, RepC3, [256]] # 16, fpn_blocks.0
  - [-1, 1, Conv, [256, 1, 1]] # 17, Y4, lateral_convs.1

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
  - [3, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 19 input_proj.0
  - [[-2, -1], 1, Concat, [1]] # cat backbone P4
  - [-1, 3, RepC3, [256]] # X3 (21), fpn_blocks.1

  - [-1, 1, Conv, [256, 3, 2]] # 22, downsample_convs.0
  - [[-1, 17], 1, Concat, [1]] # cat Y4
  - [-1, 3, RepC3, [256]] # F4 (24), pan_blocks.0

  - [-1, 1, Conv, [256, 3, 2]] # 25, downsample_convs.1
  - [[-1, 12], 1, Concat, [1]] # cat Y5
  - [-1, 3, RepC3, [256]] # F5 (27), pan_blocks.1

  - [[21, 24, 27], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)

6.2 模型改进版本2⭐

此处以 ultralytics/cfg/models/rt-detr/rtdetr-l.yaml 为例，在同目录下创建一个用于自己数据集训练的模型文件 rtdetr-l-HGBlock_ContextGuidedBlock.yaml 。

将 rtdetr-l.yaml 中的内容复制到 rtdetr-l-HGBlock_ContextGuidedBlock.yaml 文件下，修改 nc 数量等于自己数据中目标的数量。

📌 模型的修改方法是将 骨干网络 中的部分 HGBlock模块 替换成 HGBlock_ContextGuidedBlock模块 。

# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-l object detection model with P3-P5 outputs. For details see https://docs.ultralytics.com/models/rtdetr

# Parameters
nc: 1 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n-cls.yaml' will call yolov8-cls.yaml with scale 'n'
  # [depth, width, max_channels]
  l: [1.00, 1.00, 1024]

backbone:
  # [from, repeats, module, args]
  - [-1, 1, HGStem, [32, 48]] # 0-P2/4
  - [-1, 6, HGBlock, [48, 128, 3]] # stage 1

  - [-1, 1, DWConv, [128, 3, 2, 1, False]] # 2-P3/8
  - [-1, 6, HGBlock, [96, 512, 3]] # stage 2

  - [-1, 1, DWConv, [512, 3, 2, 1, False]] # 4-P4/16
  - [-1, 6, HGBlock_ContextGuidedBlock, [192, 1024, 5, True, False]] # cm, c2, k, light, shortcut
  - [-1, 6, HGBlock_ContextGuidedBlock, [192, 1024, 5, True, True]]
  - [-1, 6, HGBlock_ContextGuidedBlock, [192, 1024, 5, True, True]] # stage 3

  - [-1, 1, DWConv, [1024, 3, 2, 1, False]] # 8-P5/32
  - [-1, 6, HGBlock, [384, 2048, 5, True, False]] # stage 4

head:
  - [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 10 input_proj.2
  - [-1, 1, AIFI, [1024, 8]]
  - [-1, 1, Conv, [256, 1, 1]] # 12, Y5, lateral_convs.0

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
  - [7, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 14 input_proj.1
  - [[-2, -1], 1, Concat, [1]]
  - [-1, 3, RepC3, [256]] # 16, fpn_blocks.0
  - [-1, 1, Conv, [256, 1, 1]] # 17, Y4, lateral_convs.1

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
  - [3, 1, Conv, [256, 1, 1, None, 1, 1, False]] # 19 input_proj.0
  - [[-2, -1], 1, Concat, [1]] # cat backbone P4
  - [-1, 3, RepC3, [256]] # X3 (21), fpn_blocks.1

  - [-1, 1, Conv, [256, 3, 2]] # 22, downsample_convs.0
  - [[-1, 17], 1, Concat, [1]] # cat Y4
  - [-1, 3, RepC3, [256]] # F4 (24), pan_blocks.0

  - [-1, 1, Conv, [256, 3, 2]] # 25, downsample_convs.1
  - [[-1, 12], 1, Concat, [1]] # cat Y5
  - [-1, 3, RepC3, [256]] # F5 (27), pan_blocks.1

  - [[21, 24, 27], 1, RTDETRDecoder, [nc]] # Detect(P3, P4, P5)

---

七、成功运行结果

打印网络模型可以看到 ContextGuidedBlock_Down 和 HGBlock_ContextGuidedBlock 已经加入到模型中，并可以进行训练了。

rtdetr-l-ContextGuidedBlock_Down ：

rtdetr-l-ContextGuidedBlock_Down summary: 733 layers, 47,603,883 parameters, 47,603,883 gradients, 128.9 GFLOPs

                   from  n    params  module                                       arguments                     
  0                  -1  1     25248  ultralytics.nn.modules.block.HGStem          [3, 32, 48]                   
  1                  -1  6    155072  ultralytics.nn.modules.block.HGBlock         [48, 48, 128, 3, 6]           
  2                  -1  1    186120  ultralytics.nn.AddModules.CGNetBlock.ContextGuidedBlock_Down[128, 128]                    
  3                  -1  6    839296  ultralytics.nn.modules.block.HGBlock         [128, 96, 512, 3, 6]          
  4                  -1  1   2931744  ultralytics.nn.AddModules.CGNetBlock.ContextGuidedBlock_Down[512, 512]                    
  5                  -1  6   1695360  ultralytics.nn.modules.block.HGBlock         [512, 192, 1024, 5, 6, True, False]
  6                  -1  6   2055808  ultralytics.nn.modules.block.HGBlock         [1024, 192, 1024, 5, 6, True, True]
  7                  -1  6   2055808  ultralytics.nn.modules.block.HGBlock         [1024, 192, 1024, 5, 6, True, True]
  8                  -1  1  11696192  ultralytics.nn.AddModules.CGNetBlock.ContextGuidedBlock_Down[1024, 1024]                  
  9                  -1  6   6708480  ultralytics.nn.modules.block.HGBlock         [1024, 384, 2048, 5, 6, True, False]
 10                  -1  1    524800  ultralytics.nn.modules.conv.Conv             [2048, 256, 1, 1, None, 1, 1, False]
 11                  -1  1    789760  ultralytics.nn.modules.transformer.AIFI      [256, 1024, 8]                
 12                  -1  1     66048  ultralytics.nn.modules.conv.Conv             [256, 256, 1, 1]              
 13                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 14                   7  1    262656  ultralytics.nn.modules.conv.Conv             [1024, 256, 1, 1, None, 1, 1, False]
 15            [-2, -1]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 16                  -1  3   2232320  ultralytics.nn.modules.block.RepC3           [512, 256, 3]                 
 17                  -1  1     66048  ultralytics.nn.modules.conv.Conv             [256, 256, 1, 1]              
 18                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 19                   3  1    131584  ultralytics.nn.modules.conv.Conv             [512, 256, 1, 1, None, 1, 1, False]
 20            [-2, -1]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 21                  -1  3   2232320  ultralytics.nn.modules.block.RepC3           [512, 256, 3]                 
 22                  -1  1    590336  ultralytics.nn.modules.conv.Conv             [256, 256, 3, 2]              
 23            [-1, 17]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 24                  -1  3   2232320  ultralytics.nn.modules.block.RepC3           [512, 256, 3]                 
 25                  -1  1    590336  ultralytics.nn.modules.conv.Conv             [256, 256, 3, 2]              
 26            [-1, 12]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 27                  -1  3   2232320  ultralytics.nn.modules.block.RepC3           [512, 256, 3]                 
 28        [21, 24, 27]  1   7303907  ultralytics.nn.modules.head.RTDETRDecoder    [1, [256, 256, 256]]          
rtdetr-l-ContextGuidedBlock_Down summary: 733 layers, 47,603,883 parameters, 47,603,883 gradients, 128.9 GFLOPs

rtdetr-l-HGBlock_ContextGuidedBlock ：

rtdetr-l-HGBlock_ContextGuidedBlock summary: 739 layers, 34,818,947 parameters, 34,818,947 gradients, 113.2 GFLOPs

                   from  n    params  module                                       arguments                     
  0                  -1  1     25248  ultralytics.nn.modules.block.HGStem          [3, 32, 48]                   
  1                  -1  6    155072  ultralytics.nn.modules.block.HGBlock         [48, 48, 128, 3, 6]           
  2                  -1  1      1408  ultralytics.nn.modules.conv.DWConv           [128, 128, 3, 2, 1, False]    
  3                  -1  6    839296  ultralytics.nn.modules.block.HGBlock         [128, 96, 512, 3, 6]          
  4                  -1  1      5632  ultralytics.nn.modules.conv.DWConv           [512, 512, 3, 2, 1, False]    
  5                  -1  6   2365632  ultralytics.nn.AddModules.CGNetBlock.HGBlock_ContextGuidedBlock[512, 192, 1024, 5, 6, True, False]
  6                  -1  6   2726080  ultralytics.nn.AddModules.CGNetBlock.HGBlock_ContextGuidedBlock[1024, 192, 1024, 5, 6, True, True]
  7                  -1  6   2726080  ultralytics.nn.AddModules.CGNetBlock.HGBlock_ContextGuidedBlock[1024, 192, 1024, 5, 6, True, True]
  8                  -1  1     11264  ultralytics.nn.modules.conv.DWConv           [1024, 1024, 3, 2, 1, False]  
  9                  -1  6   6708480  ultralytics.nn.modules.block.HGBlock         [1024, 384, 2048, 5, 6, True, False]
 10                  -1  1    524800  ultralytics.nn.modules.conv.Conv             [2048, 256, 1, 1, None, 1, 1, False]
 11                  -1  1    789760  ultralytics.nn.modules.transformer.AIFI      [256, 1024, 8]                
 12                  -1  1     66048  ultralytics.nn.modules.conv.Conv             [256, 256, 1, 1]              
 13                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 14                   7  1    262656  ultralytics.nn.modules.conv.Conv             [1024, 256, 1, 1, None, 1, 1, False]
 15            [-2, -1]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 16                  -1  3   2232320  ultralytics.nn.modules.block.RepC3           [512, 256, 3]                 
 17                  -1  1     66048  ultralytics.nn.modules.conv.Conv             [256, 256, 1, 1]              
 18                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 19                   3  1    131584  ultralytics.nn.modules.conv.Conv             [512, 256, 1, 1, None, 1, 1, False]
 20            [-2, -1]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 21                  -1  3   2232320  ultralytics.nn.modules.block.RepC3           [512, 256, 3]                 
 22                  -1  1    590336  ultralytics.nn.modules.conv.Conv             [256, 256, 3, 2]              
 23            [-1, 17]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 24                  -1  3   2232320  ultralytics.nn.modules.block.RepC3           [512, 256, 3]                 
 25                  -1  1    590336  ultralytics.nn.modules.conv.Conv             [256, 256, 3, 2]              
 26            [-1, 12]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 27                  -1  3   2232320  ultralytics.nn.modules.block.RepC3           [512, 256, 3]                 
 28        [21, 24, 27]  1   7303907  ultralytics.nn.modules.head.RTDETRDecoder    [1, [256, 256, 256]]          
rtdetr-l-HGBlock_ContextGuidedBlock summary: 739 layers, 34,818,947 parameters, 34,818,947 gradients, 113.2 GFLOPs