一、本文介绍

本文改进 双HS-FPN颈部结构，融合RT-DETR中的多模态特征，以优化目标检测网络模型 。

HS-FPN 借助 通道注意力机制 及独特的 多尺度融合策略 ，有效应对 目标尺寸差异及特征稀缺问题 。针对不同模态，其 利用高级特征筛选低级特征 ，增强特征表达，助力模型精准定位和识别目标， 减少因尺度变化及特征不足导致的检测误差 ，提升 RT-DETR 在多模态检测任务中的准确性与稳定性。

---

---

二、HS-FPN介绍

Accurate Leukocyte Detection Based on Deformable-DETR and Multi-Level Feature Fusion for Aiding Diagnosis of Blood Diseases

HS - FPN结构 由 特征选择模块 和 特征融合模块 组成。

• 特征选择模块中， CA模块 先处理输入 特征图 ，经 池化 、 激活函数 确定各通道权重以 过滤特征图 ， DM模块 再对不同尺度特征图降维；
• 特征融合模块中，利用 SFF机制 ， 以高级特征为权重筛选低级特征语义信息后融合 ，提升模型检测能力。

2.1 出发点

在白细胞数据集中，白细胞识别任务面临多尺度问题，不同类型白细胞直径通常有差异，相同白细胞在不同显微镜下成像大小也会不同，这使得模型难以准确识别白细胞，所以需要设计HS - FPN来实现多尺度特征融合，帮助模型捕捉更全面的白细胞特征信息。

2.2 结构原理

• 特征选择模块 ：由 CA模块 和 DM模块 组成。对于输入特征图 $f_{in}\in R^{C\times H\times W}$ ， CA模块 先进行 全局平均池化 和全 局最大池化 ，再结合结果，经 Sigmoid激活函数 确定各通道权重 $f_{CA}\in R^{C\times 1\times 1}$ ，通过与对应尺度特征图相乘得到过滤后的特征图。因不同尺度特征图通道数不同， DM模块 用1×1卷积将各尺度特征图通道数降为 256。

• 特征融合模块 ：骨干网络生成的多尺度特征图中， 高级特征语义信息丰富但目标定位粗糙，低级特征定位精确但语义信息有限 。传统直接像素求和融合有缺陷，研究中的 SFF模块 以高级特征为权重筛选低级特征中的关键语义信息 。对于输入高级特征 $f_{high}\in R^{C\times H\times W}$ 和低级特征 $f_{low}\in R^{C\times H_1\times W_1}$ ，先对高级特征用步长为2、卷积核为3 x3的 转置卷积 扩展，再用 双线性插值 统一维度得到 $f_{att}\in R^{C\times H_1\times W_1}$ ，经 CA 模块 将高级特征转为注意力权重过滤低级特征，最后融合得到 $f_{out}\in R^{C\times H_1\times W_1}$ ，其融合过程公式为 $$f_{att}=BL\left(T-Conv\left(f_{high}\right)\right)$$ $$f_{out}=f_{low}\ast CA\left(f_{att}\right)+f_{att}$$

2.3 作用

HS-FPN 能够利用通道注意力模块，以 高级语义特征为权重过滤低级特征 ，并将筛选后的特征与高级特征逐点相加，实现多尺度特征融合，从而提高模型的特征表达能力，有助于检测到细微特征，增强模型的检测能力。

论文： https://www.sciencedirect.com/science/article/abs/pii/S0010482524000015
源码： https://github.com/JustlfC03/MFDS-DETR

三、HS-FPN的实现代码

HS-FPN模块 的实现代码如下：

import torch
import torch.nn as nn

class ChannelAttention_HSFPN(nn.Module):
    def __init__(self, in_planes, ratio=4, flag=True):
        super(ChannelAttention_HSFPN, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.max_pool = nn.AdaptiveMaxPool2d(1)

        self.conv1 = nn.Conv2d(in_planes, in_planes // ratio, 1, bias=False)
        self.relu = nn.ReLU()
        self.conv2 = nn.Conv2d(in_planes // ratio, in_planes, 1, bias=False)
        self.flag = flag
        self.sigmoid = nn.Sigmoid()

        nn.init.xavier_uniform_(self.conv1.weight)
        nn.init.xavier_uniform_(self.conv2.weight)

    def forward(self, x):
        avg_out = self.conv2(self.relu(self.conv1(self.avg_pool(x))))
        max_out = self.conv2(self.relu(self.conv1(self.max_pool(x))))
        out = avg_out + max_out
        return self.sigmoid(out) * x if self.flag else self.sigmoid(out)

class Multiply(nn.Module):
    def __init__(self) -> None:
        super().__init__()

    def forward(self, x):
        return x[0] * x[1]

class Add_HSFPN(nn.Module):
    def __init__(self):
        super().__init__()

    def forward(self, x):
        return torch.sum(torch.stack(x, dim=0), dim=0)

---

四、添加步骤

4.1 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 HSFPN.py ，将 第三节 中的代码粘贴到此处

4.2 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 .HSFPN import *

4.3 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

然后，在 parse_model函数 中添加 nn.Conv2d 、 ChannelAttention_HSFPN 、 Multiply 、 Add_HSFPN ：

 elif m is ChannelAttention_HSFPN:
     c2 = ch[f]
     args = [c2, *args]
 elif m is Multiply:
     c2 = ch[f[0]]
 elif m is Add_HSFPN:
     c2 = ch[f[-1]]

---

五、yaml模型文件

5.1 中期融合⭐

📌 此模型的修方法是将颈部网络换成HSFPN结构。

# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-ResNet50 object detection model with P3-P5 outputs.

# Parameters
ch: 6
nc: 80 # 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, IN, []]  # 0
  - [-1, 1, Multiin, [1]]  # 1
  - [-2, 1, Multiin, [2]]  # 2

  - [1, 1, ConvNormLayer, [32, 3, 2, 1, 'relu']] # 3-P1
  - [-1, 1, ConvNormLayer, [32, 3, 1, 1, 'relu']] # 4
  - [-1, 1, ConvNormLayer, [64, 3, 1, 1, 'relu']] # 5
  - [-1, 1, nn.MaxPool2d, [3, 2, 1]] # 6-P2

  - [-1, 2, Blocks, [64,  BasicBlock, 2, False]] # 7
  - [-1, 2, Blocks, [128, BasicBlock, 3, False]] # 8-P3
  - [-1, 2, Blocks, [256, BasicBlock, 4, False]] # 9-P4
  - [-1, 2, Blocks, [512, BasicBlock, 5, False]] # 10-P5

  - [2, 1, ConvNormLayer, [32, 3, 2, 1, 'relu']] # 11-P1
  - [-1, 1, ConvNormLayer, [32, 3, 1, 1, 'relu']] # 12
  - [-1, 1, ConvNormLayer, [64, 3, 1, 1, 'relu']] # 13
  - [-1, 1, nn.MaxPool2d, [3, 2, 1]] # 14-P2

  - [-1, 2, Blocks, [64,  BasicBlock, 2, False]] # 15
  - [-1, 2, Blocks, [128, BasicBlock, 3, False]] # 16-P3
  - [-1, 2, Blocks, [256, BasicBlock, 4, False]] # 17-P4
  - [-1, 2, Blocks, [512, BasicBlock, 5, False]] # 18-P5

  - [[8, 16], 1, Concat, [1]]  # 19 cat backbone P3
  - [[9, 17], 1, Concat, [1]]  # 20 cat backbone P4
  - [[10, 18], 1, Concat, [1]]  # 21 cat backbone P5

head:
  - [-1, 1, Conv, [256, 1, 1, None, 1, 1, False]]  # 22 input_proj.2
  - [-1, 1, AIFI, [1024, 8]] # 23
  - [-1, 1, Conv, [256, 1, 1]]  # 24, Y5, lateral_convs.0

  - [-1, 1, ChannelAttention_HSFPN, []] # 25
  - [-1, 1, nn.Conv2d, [256, 1]] # 26
  - [-1, 1, nn.ConvTranspose2d, [256, 3, 2, 1, 1]] # 27

  - [20, 1, ChannelAttention_HSFPN, []] # 28
  - [-1, 1, nn.Conv2d, [256, 1]] # 29
  - [27, 1, ChannelAttention_HSFPN, [4, False]] # 30
  - [[-1, -2], 1, Multiply, []] # 31
  - [[-1, 27], 1, Add, []] # 32
  - [-1, 3, RepC3, [256, 0.5]] # 33 P4/16

  - [27, 1, nn.ConvTranspose2d, [256, 3, 2, 1, 1, 16]] # 34
  - [19, 1, ChannelAttention_HSFPN, []] # 35
  - [-1, 1, nn.Conv2d, [256, 1]] # 36
  - [34, 1, ChannelAttention_HSFPN, [4, False]] # 37
  - [[-1, -2], 1, Multiply, []] # 38
  - [[-1, 34], 1, Add, []] # 39
  - [-1, 3, RepC3, [256, 0.5]] # 40 P3/8

  - [[27, 33, 40], 1, RTDETRDecoder, [nc, 256, 300, 4, 8, 3]]  # Detect(P3, P4, P5)

5.2 后期融合⭐

📌 此模型的修方法是将两个模态的颈部网络换成HSFPN结构，融合颈部部分的多模态信息。

# Ultralytics YOLO 🚀, AGPL-3.0 license
# RT-DETR-ResNet50 object detection model with P3-P5 outputs.

# Parameters
ch: 6
nc: 80 # 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, IN, []]  # 0
  - [-1, 1, Multiin, [1]]  # 1
  - [-2, 1, Multiin, [2]]  # 2

  - [1, 1, ConvNormLayer, [32, 3, 2, 1, 'relu']] # 3-P1
  - [-1, 1, ConvNormLayer, [32, 3, 1, 1, 'relu']] # 4
  - [-1, 1, ConvNormLayer, [64, 3, 1, 1, 'relu']] # 5
  - [-1, 1, nn.MaxPool2d, [3, 2, 1]] # 6-P2

  - [-1, 2, Blocks, [64,  BasicBlock, 2, False]] # 7
  - [-1, 2, Blocks, [128, BasicBlock, 3, False]] # 8-P3
  - [-1, 2, Blocks, [256, BasicBlock, 4, False]] # 9-P4
  - [-1, 2, Blocks, [512, BasicBlock, 5, False]] # 10-P5

  - [2, 1, ConvNormLayer, [32, 3, 2, 1, 'relu']] # 11-P1
  - [-1, 1, ConvNormLayer, [32, 3, 1, 1, 'relu']] # 12
  - [-1, 1, ConvNormLayer, [64, 3, 1, 1, 'relu']] # 13
  - [-1, 1, nn.MaxPool2d, [3, 2, 1]] # 14-P2

  - [-1, 2, Blocks, [64,  BasicBlock, 2, False]] # 15
  - [-1, 2, Blocks, [128, BasicBlock, 3, False]] # 16-P3
  - [-1, 2, Blocks, [256, BasicBlock, 4, False]] # 17-P4
  - [-1, 2, Blocks, [512, BasicBlock, 5, False]] # 18-P5

head:
  - [10, 1, Conv, [256, 1, 1, None, 1, 1, False]]  # 19 input_proj.2
  - [-1, 1, AIFI, [1024, 8]] # 20
  - [-1, 1, Conv, [256, 1, 1]]  # 21, Y5, lateral_convs.0

  - [-1, 1, ChannelAttention_HSFPN, []] # 22
  - [-1, 1, nn.Conv2d, [256, 1]] # 23
  - [-1, 1, nn.ConvTranspose2d, [256, 3, 2, 1, 1]] # 24

  - [9, 1, ChannelAttention_HSFPN, []] # 25
  - [-1, 1, nn.Conv2d, [256, 1]] # 26
  - [24, 1, ChannelAttention_HSFPN, [4, False]] # 27
  - [[-1, -2], 1, Multiply, []] # 28
  - [[-1, 24], 1, Add, []] # 29
  - [-1, 3, RepC3, [256, 0.5]] # 30 P4/16

  - [24, 1, nn.ConvTranspose2d, [256, 3, 2, 1, 1, 16]] # 31
  - [8, 1, ChannelAttention_HSFPN, []] # 32
  - [-1, 1, nn.Conv2d, [256, 1]] # 33
  - [31, 1, ChannelAttention_HSFPN, [4, False]] # 34
  - [[-1, -2], 1, Multiply, []] # 35
  - [[-1, 31], 1, Add, []] # 36
  - [-1, 3, RepC3, [256, 0.5]] # 37 P3/8

  - [18, 1, Conv, [256, 1, 1, None, 1, 1, False]]  # 38 input_proj.2
  - [-1, 1, AIFI, [1024, 8]] # 39
  - [-1, 1, Conv, [256, 1, 1]]  # 40, Y5, lateral_convs.0

  - [-1, 1, ChannelAttention_HSFPN, []] # 41
  - [-1, 1, nn.Conv2d, [256, 1]] # 42
  - [-1, 1, nn.ConvTranspose2d, [256, 3, 2, 1, 1]] # 43

  - [17, 1, ChannelAttention_HSFPN, []] # 44
  - [-1, 1, nn.Conv2d, [256, 1]] # 45
  - [43, 1, ChannelAttention_HSFPN, [4, False]] # 46
  - [[-1, -2], 1, Multiply, []] # 47
  - [[-1, 43], 1, Add, []] # 48
  - [-1, 3, RepC3, [256, 0.5]] # 49 P4/16

  - [43, 1, nn.ConvTranspose2d, [256, 3, 2, 1, 1, 16]] # 50
  - [16, 1, ChannelAttention_HSFPN, []] # 51
  - [-1, 1, nn.Conv2d, [256, 1]] # 52
  - [50, 1, ChannelAttention_HSFPN, [4, False]] # 53
  - [[-1, -2], 1, Multiply, []] # 54
  - [[-1, 50], 1, Add, []] # 55
  - [-1, 3, RepC3, [256, 0.5]] # 56 P3/8

  - [[24, 43], 1, Concat, [1]]  # 55 cat backbone P3
  - [[30, 49], 1, Concat, [1]]  # 56 cat backbone P4
  - [[37, 56], 1, Concat, [1]]  # 57 cat backbone P5

  - [[55, 56, 57], 1, RTDETRDecoder, [nc, 256, 300, 4, 8, 3]]  # Detect(P3, P4, P5)

---

六、成功运行结果

打印网络模型可以看到不同的融合层已经加入到模型中，并可以进行训练了。

rtdetr-resnet18-mid-HSFPN ：

rtdetr-resnet18-mid-HSFPN summary: 426 layers, 29,793,236 parameters, 29,793,236 gradients, 89.9 GFLOPs

                    from  n    params  module                                       arguments
  0                  -1  1         0  ultralytics.nn.AddModules.multimodal.IN      []
  1                  -1  1         0  ultralytics.nn.AddModules.multimodal.Multiin [1]
  2                  -2  1         0  ultralytics.nn.AddModules.multimodal.Multiin [2]
  3                   1  1       960  ultralytics.nn.AddModules.ResNet.ConvNormLayer[3, 32, 3, 2, 1, 'relu']
  4                  -1  1      9312  ultralytics.nn.AddModules.ResNet.ConvNormLayer[32, 32, 3, 1, 1, 'relu']
  5                  -1  1     18624  ultralytics.nn.AddModules.ResNet.ConvNormLayer[32, 64, 3, 1, 1, 'relu']
  6                  -1  1         0  torch.nn.modules.pooling.MaxPool2d           [3, 2, 1]
  7                  -1  2    152512  ultralytics.nn.AddModules.ResNet.Blocks      [64, 64, 2, 'BasicBlock', 2, False]
  8                  -1  2    526208  ultralytics.nn.AddModules.ResNet.Blocks      [64, 128, 2, 'BasicBlock', 3, False]
  9                  -1  2   2100992  ultralytics.nn.AddModules.ResNet.Blocks      [128, 256, 2, 'BasicBlock', 4, False]
 10                  -1  2   8396288  ultralytics.nn.AddModules.ResNet.Blocks      [256, 512, 2, 'BasicBlock', 5, False]
 11                   2  1       960  ultralytics.nn.AddModules.ResNet.ConvNormLayer[3, 32, 3, 2, 1, 'relu']
 12                  -1  1      9312  ultralytics.nn.AddModules.ResNet.ConvNormLayer[32, 32, 3, 1, 1, 'relu']
 13                  -1  1     18624  ultralytics.nn.AddModules.ResNet.ConvNormLayer[32, 64, 3, 1, 1, 'relu']
 14                  -1  1         0  torch.nn.modules.pooling.MaxPool2d           [3, 2, 1]
 15                  -1  2    152512  ultralytics.nn.AddModules.ResNet.Blocks      [64, 64, 2, 'BasicBlock', 2, False]
 16                  -1  2    526208  ultralytics.nn.AddModules.ResNet.Blocks      [64, 128, 2, 'BasicBlock', 3, False]
 17                  -1  2   2100992  ultralytics.nn.AddModules.ResNet.Blocks      [128, 256, 2, 'BasicBlock', 4, False]
 18                  -1  2   8396288  ultralytics.nn.AddModules.ResNet.Blocks      [256, 512, 2, 'BasicBlock', 5, False]
 19             [8, 16]  1         0  ultralytics.nn.modules.conv.Concat           [1]
 20             [9, 17]  1         0  ultralytics.nn.modules.conv.Concat           [1]
 21            [10, 18]  1         0  ultralytics.nn.modules.conv.Concat           [1]
 22                  -1  1    262656  ultralytics.nn.modules.conv.Conv             [1024, 256, 1, 1, None, 1, 1, False]
 23                  -1  1    789760  ultralytics.nn.modules.transformer.AIFI      [256, 1024, 8]
 24                  -1  1     66048  ultralytics.nn.modules.conv.Conv             [256, 256, 1, 1]
 25                  -1  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256]
 26                  -1  1     65792  torch.nn.modules.conv.Conv2d                 [256, 256, 1]
 27                  -1  1    590080  torch.nn.modules.conv.ConvTranspose2d        [256, 256, 3, 2, 1, 1]
 28                  20  1    131072  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[512]
 29                  -1  1    131328  torch.nn.modules.conv.Conv2d                 [512, 256, 1]
 30                  27  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256, 4, False]
 31            [-1, -2]  1         0  ultralytics.nn.AddModules.HSFPN.Multiply     []
 32            [-1, 27]  1         0  ultralytics.nn.AddModules.CFT.Add            [256]
 33                  -1  3    592384  ultralytics.nn.modules.block.RepC3           [256, 256, 3, 0.5]
 34                  27  1     37120  torch.nn.modules.conv.ConvTranspose2d        [256, 256, 3, 2, 1, 1, 16]
 35                  19  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256]
 36                  -1  1     65792  torch.nn.modules.conv.Conv2d                 [256, 256, 1]
 37                  34  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256, 4, False]
 38            [-1, -2]  1         0  ultralytics.nn.AddModules.HSFPN.Multiply     []
 39            [-1, 34]  1         0  ultralytics.nn.AddModules.CFT.Add            [256]
 40                  -1  3    592384  ultralytics.nn.modules.block.RepC3           [256, 256, 3, 0.5]
 41        [27, 33, 40]  1   3927956  ultralytics.nn.modules.head.RTDETRDecoder    [9, [256, 256, 256], 256, 300, 4, 8, 3]
rtdetr-resnet18-mid-HSFPN summary: 426 layers, 29,793,236 parameters, 29,793,236 gradients, 89.9 GFLOPs

rtdetr-resnet18-late-HSFPN ：

rtdetr-resnet18-late-HSFPN summary: 563 layers, 32,609,748 parameters, 32,609,748 gradients, 105.7 GFLOPs

                   from  n    params  module                                       arguments
  0                  -1  1         0  ultralytics.nn.AddModules.multimodal.IN      []
  1                  -1  1         0  ultralytics.nn.AddModules.multimodal.Multiin [1]
  2                  -2  1         0  ultralytics.nn.AddModules.multimodal.Multiin [2]
  3                   1  1       960  ultralytics.nn.AddModules.ResNet.ConvNormLayer[3, 32, 3, 2, 1, 'relu']
  4                  -1  1      9312  ultralytics.nn.AddModules.ResNet.ConvNormLayer[32, 32, 3, 1, 1, 'relu']
  5                  -1  1     18624  ultralytics.nn.AddModules.ResNet.ConvNormLayer[32, 64, 3, 1, 1, 'relu']
  6                  -1  1         0  torch.nn.modules.pooling.MaxPool2d           [3, 2, 1]
  7                  -1  2    152512  ultralytics.nn.AddModules.ResNet.Blocks      [64, 64, 2, 'BasicBlock', 2, False]
  8                  -1  2    526208  ultralytics.nn.AddModules.ResNet.Blocks      [64, 128, 2, 'BasicBlock', 3, False]
  9                  -1  2   2100992  ultralytics.nn.AddModules.ResNet.Blocks      [128, 256, 2, 'BasicBlock', 4, False]
 10                  -1  2   8396288  ultralytics.nn.AddModules.ResNet.Blocks      [256, 512, 2, 'BasicBlock', 5, False]
 11                   2  1       960  ultralytics.nn.AddModules.ResNet.ConvNormLayer[3, 32, 3, 2, 1, 'relu']
 12                  -1  1      9312  ultralytics.nn.AddModules.ResNet.ConvNormLayer[32, 32, 3, 1, 1, 'relu']
 13                  -1  1     18624  ultralytics.nn.AddModules.ResNet.ConvNormLayer[32, 64, 3, 1, 1, 'relu']
 14                  -1  1         0  torch.nn.modules.pooling.MaxPool2d           [3, 2, 1]
 15                  -1  2    152512  ultralytics.nn.AddModules.ResNet.Blocks      [64, 64, 2, 'BasicBlock', 2, False]
 16                  -1  2    526208  ultralytics.nn.AddModules.ResNet.Blocks      [64, 128, 2, 'BasicBlock', 3, False]
 17                  -1  2   2100992  ultralytics.nn.AddModules.ResNet.Blocks      [128, 256, 2, 'BasicBlock', 4, False]
 18                  -1  2   8396288  ultralytics.nn.AddModules.ResNet.Blocks      [256, 512, 2, 'BasicBlock', 5, False]
 19                  10  1    131584  ultralytics.nn.modules.conv.Conv             [512, 256, 1, 1, None, 1, 1, False]
 20                  -1  1    789760  ultralytics.nn.modules.transformer.AIFI      [256, 1024, 8]
 21                  -1  1     66048  ultralytics.nn.modules.conv.Conv             [256, 256, 1, 1]
 22                  -1  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256]
 23                  -1  1     65792  torch.nn.modules.conv.Conv2d                 [256, 256, 1]
 24                  -1  1    590080  torch.nn.modules.conv.ConvTranspose2d        [256, 256, 3, 2, 1, 1]
 25                   9  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256]
 26                  -1  1     65792  torch.nn.modules.conv.Conv2d                 [256, 256, 1]
 27                  24  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256, 4, False]
 28            [-1, -2]  1         0  ultralytics.nn.AddModules.HSFPN.Multiply     []
 29            [-1, 24]  1         0  ultralytics.nn.AddModules.CFT.Add            [256]
 30                  -1  3    592384  ultralytics.nn.modules.block.RepC3           [256, 256, 3, 0.5]
 31                  24  1     37120  torch.nn.modules.conv.ConvTranspose2d        [256, 256, 3, 2, 1, 1, 16]
 32                   8  1      8192  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[128]
 33                  -1  1     33024  torch.nn.modules.conv.Conv2d                 [128, 256, 1]
 34                  31  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256, 4, False]
 35            [-1, -2]  1         0  ultralytics.nn.AddModules.HSFPN.Multiply     []
 36            [-1, 31]  1         0  ultralytics.nn.AddModules.CFT.Add            [256]
 37                  -1  3    592384  ultralytics.nn.modules.block.RepC3           [256, 256, 3, 0.5]
 38                  18  1    131584  ultralytics.nn.modules.conv.Conv             [512, 256, 1, 1, None, 1, 1, False]
 39                  -1  1    789760  ultralytics.nn.modules.transformer.AIFI      [256, 1024, 8]
 40                  -1  1     66048  ultralytics.nn.modules.conv.Conv             [256, 256, 1, 1]
 41                  -1  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256]
 42                  -1  1     65792  torch.nn.modules.conv.Conv2d                 [256, 256, 1]
 43                  -1  1    590080  torch.nn.modules.conv.ConvTranspose2d        [256, 256, 3, 2, 1, 1]
 44                  17  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256]
 45                  -1  1     65792  torch.nn.modules.conv.Conv2d                 [256, 256, 1]
 46                  43  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256, 4, False]
 47            [-1, -2]  1         0  ultralytics.nn.AddModules.HSFPN.Multiply     []
 48            [-1, 43]  1         0  ultralytics.nn.AddModules.CFT.Add            [256]
 49                  -1  3    592384  ultralytics.nn.modules.block.RepC3           [256, 256, 3, 0.5]
 50                  43  1     37120  torch.nn.modules.conv.ConvTranspose2d        [256, 256, 3, 2, 1, 1, 16]
 51                  16  1      8192  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[128]
 52                  -1  1     33024  torch.nn.modules.conv.Conv2d                 [128, 256, 1]
 53                  50  1     32768  ultralytics.nn.AddModules.HSFPN.ChannelAttention_HSFPN[256, 4, False]
 54            [-1, -2]  1         0  ultralytics.nn.AddModules.HSFPN.Multiply     []
 55            [-1, 50]  1         0  ultralytics.nn.AddModules.CFT.Add            [256]
 56                  -1  3    592384  ultralytics.nn.modules.block.RepC3           [256, 256, 3, 0.5]
 57            [24, 43]  1         0  ultralytics.nn.modules.conv.Concat           [1]
 58            [30, 49]  1         0  ultralytics.nn.modules.conv.Concat           [1]
 59            [37, 56]  1         0  ultralytics.nn.modules.conv.Concat           [1]
 60        [55, 56, 57]  1   3993492  ultralytics.nn.modules.head.RTDETRDecoder    [9, [256, 256, 512], 256, 300, 4, 8, 3]
rtdetr-resnet18-late-HSFPN summary: 563 layers, 32,609,748 parameters, 32,609,748 gradients, 105.7 GFLOPs