• Yolov8有效涨点：YOLOv8-AM，添加多种注意力模块提高检测精度，含代码，超详细

    前言

    2023 年，Ultralytics 推出了最新版本的 YOLO 模型。注意力机制是提高模型性能最热门的方法之一。

    本次介绍的是YOLOv8-AM，它将注意力机制融入到原始的YOLOv8架构中。具体来说，我们分别采用四个注意力模块：卷积块注意力模块（CBAM）、全局注意力机制（GAM）、高效通道注意力（ECA）和随机注意力（SA）来设计改进模型并在数据集上进行测试。实验结果表明，基于ResBlock + CBAM（ResCBAM）的YOLOv8-AM模型在IoU 50（mAP 50）下的平均精度提到了2.2%，达到了state-of-the-art（SOTA）表现。相反，结合GAM的YOLOv8-AM模型获得了的mAP @50并不是一个令人满意的增强。因此，我们将ResBlock和GAM结合起来，引入ResGAM设计另一个新的YOLOv8-AM模型，获得一个较为满意的结果。

    目录

    前言

    注意力机制：

    Convolutional Block Attention Module

    Efficient Channel Attention

    Shuffle Attention

    Global Attention Mechanism

    实验结果（供参考）

    可论文指导--------->v jiabei-545

    改进代码(失效+ v 👆)

    注意力机制：
    带有YOLOv8-AM的结构图

    YOLOv8 架构由四个关键组件组成：Backbone、Neck、Head 和 Loss Function。 Backbone 融合了 Cross Stage Partial (CSP) 概念，具有减少计算负载、同时增强 CNN 学习能力的优势。如图所示，YOLOv8与采用C3模块的YOLOv5不同，采用C2f模块，该模块集成了C3模块和YOLOv7中的扩展ELAN（E-ELAN）概念。

    YOLOv8-AM模型架构详解，其中注意力模块为Shuffle Attention（SA）、Efficient Channel Attention（ECA）、Global Attention Mechanism（GAM）、ResBlock + Convolutional Block Attention Module（ResCBAM）

    Convolutional Block Attention Module
    CBAM架构

    CBAM 包括通道注意力（C-Attention）和空间注意力（S-Attention），如图所示。给定一个中间特征图，CBAM 通过等式依次推断出 1D 通道注意力图  和 2D 空间注意力图 。

    ResBlock + Convolutional Block Attention Module

    原理和resnet一样 

    1. # Ultralytics YOLO 🚀, AGPL-3.0 license
    2. # YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
    3.  
    4. # Parameters
    5. nc: 9  # number of classes
    6. scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
    7.   # [depth, width, max_channels]
    8.   n: [0.33, 0.25, 1024]  # YOLOv8n summary: 225 layers,  3157200 parameters,  3157184 gradients,   8.9 GFLOPs
    9.   s: [0.33, 0.50, 1024]  # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients,  28.8 GFLOPs
    10.   m: [0.67, 0.75, 768]   # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients,  79.3 GFLOPs
    11.   l: [1.00, 1.00, 512]   # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
    12.   x: [1.00, 1.25, 512]   # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
    13.  
    14. # YOLOv8.0n backbone
    15. backbone:
    16.   # [from, repeats, module, args]
    17.   - [-1, 1, Conv, [64, 3, 2]]  # 0-P1/2
    18.   - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
    19.   - [-1, 3, C2f, [128, True]]
    20.   - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
    21.   - [-1, 6, C2f, [256, True]]
    22.   - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
    23.   - [-1, 6, C2f, [512, True]]
    24.   - [-1, 1, Conv, [1024, 3, 2]]  # 7-P5/32
    25.   - [-1, 3, C2f, [1024, True]]
    26.   - [-1, 1, SPPF, [1024, 5]]  # 9
    27.  
    28. # YOLOv8.0n head
    29. head:
    30.   - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
    31.   - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
    32.   - [-1, 3, C2f, [512]]  # 12
    33.   - [-1, 1, ResBlock_CBAM, [512]]
    34.  
    35.   - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
    36.   - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
    37.   - [-1, 3, C2f, [256]]  # 16 (P3/8-small)
    38.   - [-1, 1, ResBlock_CBAM, [256]]
    39.  
    40.   - [-1, 1, Conv, [256, 3, 2]]
    41.   - [[-1, 12], 1, Concat, [1]]  # cat head P4
    42.   - [-1, 3, C2f, [512]]  # 20 (P4/16-medium)
    43.   - [-1, 1, ResBlock_CBAM, [512]]
    44.  
    45.   - [-1, 1, Conv, [512, 3, 2]]
    46.   - [[-1, 9], 1, Concat, [1]]  # cat head P5
    47.   - [-1, 3, C2f, [1024]]  # 24 (P5/32-large)
    48.   - [-1, 1, ResBlock_CBAM, [1024]]
    49.  
    50.   - [[17, 21, 25], 1, Detect, [nc]]  # Detect(P3, P4, P5)
    Efficient Channel Attention
    Efficient Channel Attention

    ECA 主要包含跨通道交互和具有自适应卷积核的一维卷积，如图 所示。跨通道交互代表了一种组合特征的新方法，增强了特定语义的特征表达。

    1. # Ultralytics YOLO 🚀, AGPL-3.0 license
    2. # YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
    3.  
    4. # Parameters
    5. nc: 9  # number of classes
    6. scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
    7.   # [depth, width, max_channels]
    8.   n: [0.33, 0.25, 1024]  # YOLOv8n summary: 225 layers,  3157200 parameters,  3157184 gradients,   8.9 GFLOPs
    9.   s: [0.33, 0.50, 1024]  # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients,  28.8 GFLOPs
    10.   m: [0.67, 0.75, 768]   # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients,  79.3 GFLOPs
    11.   l: [1.00, 1.00, 512]   # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
    12.   x: [1.00, 1.25, 512]   # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
    13.  
    14. # YOLOv8.0n backbone
    15. backbone:
    16.   # [from, repeats, module, args]
    17.   - [-1, 1, Conv, [64, 3, 2]]  # 0-P1/2
    18.   - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
    19.   - [-1, 3, C2f, [128, True]]
    20.   - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
    21.   - [-1, 6, C2f, [256, True]]
    22.   - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
    23.   - [-1, 6, C2f, [512, True]]
    24.   - [-1, 1, Conv, [1024, 3, 2]]  # 7-P5/32
    25.   - [-1, 3, C2f, [1024, True]]
    26.   - [-1, 1, SPPF, [1024, 5]]  # 9
    27.  
    28. # YOLOv8.0n head
    29. head:
    30.   - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
    31.   - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
    32.   - [-1, 3, C2f, [512]]  # 12
    33.   - [-1, 1, ECAAttention, [512]]
    34.  
    35.   - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
    36.   - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
    37.   - [-1, 3, C2f, [256]]  # 16 (P3/8-small)
    38.   - [-1, 1, ECAAttention, [256]]
    39.  
    40.   - [-1, 1, Conv, [256, 3, 2]]
    41.   - [[-1, 12], 1, Concat, [1]]  # cat head P4
    42.   - [-1, 3, C2f, [512]]  # 20 (P4/16-medium)
    43.   - [-1, 1, ECAAttention, [512]]
    44.  
    45.   - [-1, 1, Conv, [512, 3, 2]]
    46.   - [[-1, 9], 1, Concat, [1]]  # cat head P5
    47.   - [-1, 3, C2f, [1024]]  # 24 (P5/32-large)
    48.   - [-1, 1, ECAAttention, [1024]]
    49.  
    50.   - [[17, 21, 25], 1, Detect, [nc]]  # Detect(P3, P4, P5)
    Shuffle Attention
    Shuffle Attention

     SA将输入特征图分为不同的组，利用Shuffle Unit将通道注意力和空间注意力整合到每个组的一个块中，如图所示。随后，子特征被聚合，并且“ ShuffleNetV2 中使用的“Channel Shuffle”算子用于促进各种子特征之间的信息通信。对于通道注意力，SA 采用 GAP 来捕获和嵌入子特征。此外，使用带有 sigmoid 函数的简单门控机制来创建紧凑的函数，以促进精确和自适应的选择。

    1. # SA.yaml
    2. # Ultralytics YOLO 🚀, AGPL-3.0 license
    3. # YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
    4.  
    5. # Parameters
    6. nc: 9  # number of classes
    7. scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
    8.   # [depth, width, max_channels]
    9.   n: [0.33, 0.25, 1024]  # YOLOv8n summary: 225 layers,  3157200 parameters,  3157184 gradients,   8.9 GFLOPs
    10.   s: [0.33, 0.50, 1024]  # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients,  28.8 GFLOPs
    11.   m: [0.67, 0.75, 768]   # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients,  79.3 GFLOPs
    12.   l: [1.00, 1.00, 512]   # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
    13.   x: [1.00, 1.25, 512]   # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
    14.  
    15. # YOLOv8.0n backbone
    16. backbone:
    17.   # [from, repeats, module, args]
    18.   - [-1, 1, Conv, [64, 3, 2]]  # 0-P1/2
    19.   - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
    20.   - [-1, 3, C2f, [128, True]]
    21.   - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
    22.   - [-1, 6, C2f, [256, True]]
    23.   - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
    24.   - [-1, 6, C2f, [512, True]]
    25.   - [-1, 1, Conv, [1024, 3, 2]]  # 7-P5/32
    26.   - [-1, 3, C2f, [1024, True]]
    27.   - [-1, 1, SPPF, [1024, 5]]  # 9
    28.  
    29. # YOLOv8.0n head
    30. head:
    31.   - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
    32.   - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
    33.   - [-1, 3, C2f, [512]]  # 12
    34.   - [-1, 1, ShuffleAttention, [512]]
    35.  
    36.   - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
    37.   - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
    38.   - [-1, 3, C2f, [256]]  # 16 (P3/8-small)
    39.   - [-1, 1, ShuffleAttention, [256]]
    40.  
    41.   - [-1, 1, Conv, [256, 3, 2]]
    42.   - [[-1, 12], 1, Concat, [1]]  # cat head P4
    43.   - [-1, 3, C2f, [512]]  # 20 (P4/16-medium)
    44.   - [-1, 1, ShuffleAttention, [512]]
    45.  
    46.   - [-1, 1, Conv, [512, 3, 2]]
    47.   - [[-1, 9], 1, Concat, [1]]  # cat head P5
    48.   - [-1, 3, C2f, [1024]]  # 24 (P5/32-large)
    49.   - [-1, 1, ShuffleAttention, [1024]]
    50.  
    51.   - [[17, 21, 25], 1, Detect, [nc]]  # Detect(P3, P4, P5)
    Global Attention Mechanism
    Global Attention Mechanism

     GAM采用了CBAM提出的由通道注意力和空间注意力组成的主要架构，并重新设计了子模块，如图所示。此外，我在GAM内的各层之间添加了快捷连接，这使得输入能够更快地向前传播。

    1. # GAM.yaml
    2. # Ultralytics YOLO 🚀, AGPL-3.0 license
    3. # YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
    4.  
    5. # Parameters
    6. nc: 9  # number of classes
    7. scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
    8.   # [depth, width, max_channels]
    9.   n: [0.33, 0.25, 1024]  # YOLOv8n summary: 225 layers,  3157200 parameters,  3157184 gradients,   8.9 GFLOPs
    10.   s: [0.33, 0.50, 1024]  # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients,  28.8 GFLOPs
    11.   m: [0.67, 0.75, 768]   # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients,  79.3 GFLOPs
    12.   l: [1.00, 1.00, 512]   # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
    13.   x: [1.00, 1.25, 512]   # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
    14.  
    15. # YOLOv8.0n backbone
    16. backbone:
    17.   # [from, repeats, module, args]
    18.   - [-1, 1, Conv, [64, 3, 2]]  # 0-P1/2
    19.   - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
    20.   - [-1, 3, C2f, [128, True]]
    21.   - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
    22.   - [-1, 6, C2f, [256, True]]
    23.   - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
    24.   - [-1, 6, C2f, [512, True]]
    25.   - [-1, 1, Conv, [1024, 3, 2]]  # 7-P5/32
    26.   - [-1, 3, C2f, [1024, True]]
    27.   - [-1, 1, SPPF, [1024, 5]]  # 9
    28.  
    29. # YOLOv8.0n head
    30. head:
    31.   - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
    32.   - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
    33.   - [-1, 3, C2f, [512]]  # 12
    34.   - [-1, 1, GAM_Attention, [512,512]]
    35.  
    36.   - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
    37.   - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
    38.   - [-1, 3, C2f, [256]]  # 16 (P3/8-small)
    39.   - [-1, 1, GAM_Attention, [256,256]]
    40.  
    41.   - [-1, 1, Conv, [256, 3, 2]]
    42.   - [[-1, 12], 1, Concat, [1]]  # cat head P4
    43.   - [-1, 3, C2f, [512]]  # 20 (P4/16-medium)
    44.   - [-1, 1, GAM_Attention, [512,512]]
    45.  
    46.   - [-1, 1, Conv, [512, 3, 2]]
    47.   - [[-1, 9], 1, Concat, [1]]  # cat head P5
    48.   - [-1, 3, C2f, [1024]]  # 24 (P5/32-large)
    49.   - [-1, 1, GAM_Attention, [1024,1024]]
    50.  
    51.   - [[17, 21, 25], 1, Detect, [nc]]  # Detect(P3, P4, P5)
    ResBlock + Global Attention Mechanism

     原理和resnet一样

    实验结果（供参考）

    ResBlock + Convolutional Block Attention Module
    Shuffle Attention

    Efficient Channel Attention
    Global Attention Mechanism

    ResBlock + Global Attention Mechanism
    定量比较（Precision/Recall/F1/mAP）

    可论文指导--------->v jiabei-545
    改进代码(失效+ v 👆)

    链接: https://pan.baidu.com/s/1Fi7ghwJ6XiXrDDnoCvlvrQ?pwd=zk88 提取码: zk88 

    欢迎大家在评论区进行讨论

  • 相关阅读:
    发明专利一般多长时间受理
    C++宏的用法
    什么是周转时间？
    国密算法：利用python进行sm2非对称算法的实现，国密算法库gmssl的使用
    Spring Boot的创建和使用（JavaEE进阶系列2）
    半小时速通Python爬虫！GitHub开源的Python爬虫入门教程
    SpringSecurity初探(一)
    java基于ssm的电子资源管理系统 毕业设计
    Java 多线程写zip文件遇到的错误 write beyond end of stream！
    基于区块链的隐私保护数据共享
  • 原文地址：https://blog.csdn.net/m0_67647321/article/details/136281570