YOLO26改进| Neck | 轻量化且有效提升小目标检测效果

kay_545

833人浏览 · 2026-05-07 10:21:41

kay_545 · 2026-05-07 10:21:41 发布

💡💡💡本专栏所有程序均经过测试，可成功执行💡💡💡

本文给大家带来的教程是将YOLO26的neck替换为Slim-Neck来提取特征。文章在介绍主要的原理后，将手把手教学如何进行模块的代码添加和修改，并将修改后的完整代码放在文章的最后，方便大家一键运行，小白也可轻松上手实践。以帮助您更好地学习深度学习目标检测YOLO系列的挑战。

专栏地址：YOLO26改进-论文涨点——点击跳转看所有内容，关注不迷路！

1.论文

2. Slim-Neck代码实现

2.1 将Slim-Neck添加到YOLO26中

1.论文

论文地址：Slim-neck by GSConv: A lightweight-design for real-time detector architectures

官方代码：官方代码仓库点击即可跳转

2. Slim-Neck代码实现

2.1 将Slim-Neck添加到YOLO26中

关键步骤一：在ultralytics\ultralytics\nn\modules下面新建文件夹models，在文件夹下新建Slim_Neck.py，粘贴下面代码


 
import torch
import torch.nn as nn 

from ultralytics.nn.modules.conv import Conv, DWConv    

__all__ = ("GSConv", "GSBottleneck", "GSBottleneckC", "VoVGSCSP")


class GSConv(nn.Module):
    """GSConv from Slim-neck by GSConv."""

    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, d=1, act=True):  
        super().__init__()
        c_ = c2 // 2     
        self.cv1 = Conv(c1, c_, k, s, p, g, d, act)
        self.cv2 = Conv(c_, c_, 5, 1, p, c_, d, act)

    def forward(self, x):
        x1 = self.cv1(x)
        x2 = torch.cat((x1, self.cv2(x1)), 1)    
 
        b, n, h, w = x2.size()
        b_n = b * n // 2 
        y = x2.reshape(b_n, 2, h * w)   
        y = y.permute(1, 0, 2)
        y = y.reshape(2, -1, n // 2, h, w)
        return torch.cat((y[0], y[1]), 1)
    
     
class GSBottleneck(nn.Module):
    """GS bottleneck used by SlimNeck CSP blocks.""" 

    def __init__(self, c1, c2, k=3, s=1, e=0.5): 
        super().__init__()
        c_ = int(c2 * e)
        self.conv_lighting = nn.Sequential(GSConv(c1, c_, 1, 1), GSConv(c_, c2, 3, 1, act=False))
        self.shortcut = Conv(c1, c2, 1, 1, act=False)  

    def forward(self, x):
        return self.conv_lighting(x) + self.shortcut(x)

     
class GSBottleneckC(GSBottleneck): 
    """Cheap GS bottleneck variant with depthwise shortcut."""

    def __init__(self, c1, c2, k=3, s=1):
        super().__init__(c1, c2, k, s)
        self.shortcut = DWConv(c1, c2, k, s, act=False)   
    

class VoVGSCSP(nn.Module):     
    """VoV-GSCSP neck block composed of GS bottlenecks."""
  
    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        super().__init__()
        c_ = int(c2 * e)
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c1, c_, 1, 1)  
        self.gsb = nn.Sequential(*(GSBottleneck(c_, c_, e=1.0) for _ in range(n)))   
        # Kept for source-state compatibility with the upstream SlimNeck implementation.     
        self.res = Conv(c_, c_, 3, 1, act=False)
        self.cv3 = Conv(2 * c_, c2, 1)

    def forward(self, x):     
        x1 = self.gsb(self.cv1(x)) 
        y = self.cv2(x)    
        return self.cv3(torch.cat((y, x1), dim=1))

2.2 更改init.py文件

关键步骤二：在文件ultralytics\ultralytics\nn\modules\models文件夹下新建__init__.py文件，先导入函数

然后在下面的__all__中声明函数

2.3 添加yaml文件

关键步骤三：在/ultralytics/ultralytics/cfg/models/26下面新建文件yolo26_Slim_Neck.yaml文件，粘贴下面的内容

目标检测

# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license

# Ultralytics YOLO26 object detection model with P3/8 - P5/32 outputs
# Model docs: https://docs.ultralytics.com/models/yolo26
# Task docs: https://docs.ultralytics.com/tasks/detect

# Parameters
nc: 80 # number of classes
end2end: True # whether to use end-to-end mode
reg_max: 1 # DFL bins
scales: # model compound scaling constants, i.e. 'model=yolo26n.yaml' will call yolo26.yaml with scale 'n'
  # [depth, width, max_channels]
  n: [0.50, 0.25, 1024] # summary: 260 layers, 2,572,280 parameters, 2,572,280 gradients, 6.1 GFLOPs
  s: [0.50, 0.50, 1024] # summary: 260 layers, 10,009,784 parameters, 10,009,784 gradients, 22.8 GFLOPs
  m: [0.50, 1.00, 512] # summary: 280 layers, 21,896,248 parameters, 21,896,248 gradients, 75.4 GFLOPs
  l: [1.00, 1.00, 512] # summary: 392 layers, 26,299,704 parameters, 26,299,704 gradients, 93.8 GFLOPs
  x: [1.00, 1.50, 512] # summary: 392 layers, 58,993,368 parameters, 58,993,368 gradients, 209.5 GFLOPs

backbone:
  - [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
  - [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
  - [-1, 2, C3k2, [256, False, 0.25]] # 2-P2/4
  - [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
  - [-1, 2, C3k2, [512, False, 0.25]] # 4-P3/8
  - [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
  - [-1, 2, C3k2, [512, True]] # 6-P4/16
  - [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
  - [-1, 2, C3k2, [1024, True]] # 8-P5/32
  - [-1, 1, SPPF, [1024, 5, 3, True]] # 9-P5/32
  - [-1, 2, C2PSA, [1024]] # 10-P5/32

head:
  - [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 11-P4/16
  - [[-1, 6], 1, Concat, [1]] # 12-P4/16
  - [-1, 2, VoVGSCSP, [512]] # 13-P4/16

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 14-P3/8
  - [[-1, 4], 1, Concat, [1]] # 15-P3/8
  - [-1, 2, VoVGSCSP, [256]] # 16-P3/8

  - [-1, 1, GSConv, [256, 3, 2]] # 17-P4/16
  - [[-1, 13], 1, Concat, [1]] # 18-P4/16
  - [-1, 2, VoVGSCSP, [512]] # 19-P4/16

  - [-1, 1, GSConv, [512, 3, 2]] # 20-P5/32
  - [[-1, 10], 1, Concat, [1]] # 21-P5/32
  - [-1, 1, VoVGSCSP, [1024]] # 22-P5/32

  - [[16, 19, 22], 1, Detect, [nc]] # 23-P3/8,P4/16,P5/32

语义分割

# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license

# Ultralytics YOLO26 object detection model with P3/8 - P5/32 outputs
# Model docs: https://docs.ultralytics.com/models/yolo26
# Task docs: https://docs.ultralytics.com/tasks/detect

# Parameters
nc: 80 # number of classes
end2end: True # whether to use end-to-end mode
reg_max: 1 # DFL bins
scales: # model compound scaling constants, i.e. 'model=yolo26n.yaml' will call yolo26.yaml with scale 'n'
  # [depth, width, max_channels]
  n: [0.50, 0.25, 1024] # summary: 260 layers, 2,572,280 parameters, 2,572,280 gradients, 6.1 GFLOPs
  s: [0.50, 0.50, 1024] # summary: 260 layers, 10,009,784 parameters, 10,009,784 gradients, 22.8 GFLOPs
  m: [0.50, 1.00, 512] # summary: 280 layers, 21,896,248 parameters, 21,896,248 gradients, 75.4 GFLOPs
  l: [1.00, 1.00, 512] # summary: 392 layers, 26,299,704 parameters, 26,299,704 gradients, 93.8 GFLOPs
  x: [1.00, 1.50, 512] # summary: 392 layers, 58,993,368 parameters, 58,993,368 gradients, 209.5 GFLOPs

backbone:
  - [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
  - [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
  - [-1, 2, C3k2, [256, False, 0.25]] # 2-P2/4
  - [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
  - [-1, 2, C3k2, [512, False, 0.25]] # 4-P3/8
  - [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
  - [-1, 2, C3k2, [512, True]] # 6-P4/16
  - [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
  - [-1, 2, C3k2, [1024, True]] # 8-P5/32
  - [-1, 1, SPPF, [1024, 5, 3, True]] # 9-P5/32
  - [-1, 2, C2PSA, [1024]] # 10-P5/32

head:
  - [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 11-P4/16
  - [[-1, 6], 1, Concat, [1]] # 12-P4/16
  - [-1, 2, VoVGSCSP, [512]] # 13-P4/16

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 14-P3/8
  - [[-1, 4], 1, Concat, [1]] # 15-P3/8
  - [-1, 2, VoVGSCSP, [256]] # 16-P3/8

  - [-1, 1, GSConv, [256, 3, 2]] # 17-P4/16
  - [[-1, 13], 1, Concat, [1]] # 18-P4/16
  - [-1, 2, VoVGSCSP, [512]] # 19-P4/16

  - [-1, 1, GSConv, [512, 3, 2]] # 20-P5/32
  - [[-1, 10], 1, Concat, [1]] # 21-P5/32
  - [-1, 1, VoVGSCSP, [1024]] # 22-P5/32

  - [[16, 19, 22], 1, Segment, [nc, 32, 256]]

旋转目标检测

# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license

# Ultralytics YOLO26 object detection model with P3/8 - P5/32 outputs
# Model docs: https://docs.ultralytics.com/models/yolo26
# Task docs: https://docs.ultralytics.com/tasks/detect

# Parameters
nc: 80 # number of classes
end2end: True # whether to use end-to-end mode
reg_max: 1 # DFL bins
scales: # model compound scaling constants, i.e. 'model=yolo26n.yaml' will call yolo26.yaml with scale 'n'
  # [depth, width, max_channels]
  n: [0.50, 0.25, 1024] # summary: 260 layers, 2,572,280 parameters, 2,572,280 gradients, 6.1 GFLOPs
  s: [0.50, 0.50, 1024] # summary: 260 layers, 10,009,784 parameters, 10,009,784 gradients, 22.8 GFLOPs
  m: [0.50, 1.00, 512] # summary: 280 layers, 21,896,248 parameters, 21,896,248 gradients, 75.4 GFLOPs
  l: [1.00, 1.00, 512] # summary: 392 layers, 26,299,704 parameters, 26,299,704 gradients, 93.8 GFLOPs
  x: [1.00, 1.50, 512] # summary: 392 layers, 58,993,368 parameters, 58,993,368 gradients, 209.5 GFLOPs

backbone:
  - [-1, 1, Conv, [64, 3, 2]] # 0-P1/2
  - [-1, 1, Conv, [128, 3, 2]] # 1-P2/4
  - [-1, 2, C3k2, [256, False, 0.25]] # 2-P2/4
  - [-1, 1, Conv, [256, 3, 2]] # 3-P3/8
  - [-1, 2, C3k2, [512, False, 0.25]] # 4-P3/8
  - [-1, 1, Conv, [512, 3, 2]] # 5-P4/16
  - [-1, 2, C3k2, [512, True]] # 6-P4/16
  - [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32
  - [-1, 2, C3k2, [1024, True]] # 8-P5/32
  - [-1, 1, SPPF, [1024, 5, 3, True]] # 9-P5/32
  - [-1, 2, C2PSA, [1024]] # 10-P5/32

head:
  - [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 11-P4/16
  - [[-1, 6], 1, Concat, [1]] # 12-P4/16
  - [-1, 2, VoVGSCSP, [512]] # 13-P4/16

  - [-1, 1, nn.Upsample, [None, 2, "nearest"]] # 14-P3/8
  - [[-1, 4], 1, Concat, [1]] # 15-P3/8
  - [-1, 2, VoVGSCSP, [256]] # 16-P3/8

  - [-1, 1, GSConv, [256, 3, 2]] # 17-P4/16
  - [[-1, 13], 1, Concat, [1]] # 18-P4/16
  - [-1, 2, VoVGSCSP, [512]] # 19-P4/16

  - [-1, 1, GSConv, [512, 3, 2]] # 20-P5/32
  - [[-1, 10], 1, Concat, [1]] # 21-P5/32
  - [-1, 1, VoVGSCSP, [1024]] # 22-P5/32

  - [[16, 19, 22], 1, OBB, [nc, 1]]

温馨提示：本文只是对yolo26基础上添加模块，如果要对yolo26 n/l/m/x进行添加则只需要指定对应的depth_multiple 和 width_multiple

end2end: True # whether to use end-to-end mode
reg_max: 1 # DFL bins
scales: # model compound scaling constants, i.e. 'model=yolo26n.yaml' will call yolo26.yaml with scale 'n'
  # [depth, width, max_channels]
  n: [0.50, 0.25, 1024] # summary: 260 layers, 2,572,280 parameters, 2,572,280 gradients, 6.1 GFLOPs
  s: [0.50, 0.50, 1024] # summary: 260 layers, 10,009,784 parameters, 10,009,784 gradients, 22.8 GFLOPs
  m: [0.50, 1.00, 512] # summary: 280 layers, 21,896,248 parameters, 21,896,248 gradients, 75.4 GFLOPs
  l: [1.00, 1.00, 512] # summary: 392 layers, 26,299,704 parameters, 26,299,704 gradients, 93.8 GFLOPs
  x: [1.00, 1.50, 512] # summary: 392 layers, 58,993,368 parameters, 58,993,368 gradients, 209.5 GFLOPs

2.4 在task.py中进行注册

关键步骤四：在parse_model函数中进行注册，添加Slim_Neck

先在task.py导入函数

然后在task.py文件下找到parse_model这个函数，如下图，添加Slim_Neck

2.5 执行程序

关键步骤五: 在ultralytics文件中新建train.py，将model的参数路径设置为yolo26_Slim_Neck.yaml的路径即可【注意是在外边的Ultralytics下新建train.py】

from ultralytics import YOLO
import warnings
warnings.filterwarnings('ignore')
from pathlib import Path
 
if __name__ == '__main__':
 
 
    # 加载模型
    model = YOLO("ultralytics/cfg/26/yolo26.yaml")  # 你要选择的模型yaml文件地址
    # Use the model
    results = model.train(data=r"你的数据集的yaml文件地址",
                          epochs=100, batch=16, imgsz=640, workers=4, name=Path(model.cfg).stem)  # 训练模型

🚀运行程序，如果出现下面的内容则说明添加成功🚀

                   from  n    params  module                                       arguments                     
  0                  -1  1       464  ultralytics.nn.modules.conv.Conv             [3, 16, 3, 2]                 
  1                  -1  1      4672  ultralytics.nn.modules.conv.Conv             [16, 32, 3, 2]                
  2                  -1  1      6640  ultralytics.nn.modules.block.C3k2            [32, 64, 1, False, 0.25]      
  3                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
  4                  -1  1     26080  ultralytics.nn.modules.block.C3k2            [64, 128, 1, False, 0.25]     
  5                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
  6                  -1  1     87040  ultralytics.nn.modules.block.C3k2            [128, 128, 1, True]           
  7                  -1  1    295424  ultralytics.nn.modules.conv.Conv             [128, 256, 3, 2]              
  8                  -1  1    346112  ultralytics.nn.modules.block.C3k2            [256, 256, 1, True]           
  9                  -1  1    164608  ultralytics.nn.modules.block.SPPF            [256, 256, 5, 3, True]        
 10                  -1  1    249728  ultralytics.nn.modules.block.C2PSA           [256, 256, 1]                 
 11                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 12             [-1, 6]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 13                  -1  1    129600  ultralytics.nn.models.Slim_Neck.VoVGSCSP     [384, 128, 1]                 
 14                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 15             [-1, 4]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 16                  -1  1     37152  ultralytics.nn.models.Slim_Neck.VoVGSCSP     [256, 64, 1]                  
 17                  -1  1     19360  ultralytics.nn.models.Slim_Neck.GSConv       [64, 64, 3, 2]                
 18            [-1, 13]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 19                  -1  1    105024  ultralytics.nn.models.Slim_Neck.VoVGSCSP     [192, 128, 1]                 
 20                  -1  1     75584  ultralytics.nn.models.Slim_Neck.GSConv       [128, 128, 3, 2]              
 21            [-1, 10]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 22                  -1  1    414848  ultralytics.nn.models.Slim_Neck.VoVGSCSP     [384, 256, 1]                 
 23        [16, 19, 22]  1    309656  ultralytics.nn.modules.head.Detect           [80, 1, True, [64, 128, 256]] 
YOLO26_Slim_Neck summary: 276 layers, 2,456,696 parameters, 2,456,696 gradients, 5.6 GFLOPs

3. 完整代码分享

主页侧边

4. GFLOPs

关于GFLOPs的计算方式可以查看：百面算法工程师 | 卷积基础知识——Convolution

未改进的YOLO26n GFLOPs

改进后的GFLOPs

5. 进阶

可以与其他的注意力机制或者损失函数等结合，进一步提升检测效果

6.总结

通过以上的改进方法，我们成功提升了模型的表现。这只是一个开始，未来还有更多优化和技术深挖的空间。在这里，我想隆重向大家推荐我的专栏——<专栏地址：YOLO26改进-论文涨点——点击跳转看所有内容，关注不迷路！>。这个专栏专注于前沿的深度学习技术，特别是目标检测领域的最新进展，不仅包含对YOLO26的深入解析和改进策略，还会定期更新来自各大顶会（如CVPR、NeurIPS等）的论文复现和实战分享。

为什么订阅我的专栏？ ——专栏地址：YOLO26改进-论文涨点——点击跳转看所有内容，关注不迷路！