Skip to content

自用改进模块-Self-Use Improvement Module

License

Notifications You must be signed in to change notification settings

GuGuGuGun/model

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

25 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

自用改进模块仓库

Self-use Improved Model project

License


📝 Table of Contents

🧐 About

  • 这是一个自用仓库,主要方向CV,主要目的是为了保存自己修改的各种模块改进和各种Net的改进 有时候也会上传其他东西
  • This is a self-use repository, the main direction of CV, the main purpose is to save the various module improvements and various Net improvements that you have modified, and sometimes upload other things

🎈 Model

  • C2F方向改进:

    1、CCE-一种融合了ELA注意力、以CCA重新设计bottleneck、使用CGAFusion进行双尺度特征融合的C2f改进模块(在NEU-DET数据集上表现良好)

    CCE-an improved C2f module that fuses ELA attention module, redesigns bottleneck with CCA, and uses CGAFusion for dual-scale feature fusion (performs well on NEU-DET dataset)

    2、star-C2f-使用StarNet的block替换C2f的bottleneck,更加轻量化的C2f改进模块

    star-C2f - Replace the bottleneck of C2f with StarNet's block, and improve the module of C2f with lightweight

    3、CPCA-C2f-使用CPCA注意力模块与C2f的bottleneck融合,增强空间关系的提取能力,提高特征的表征能力

    CPCA-C2f - The bottleneck fusion of CPCA attention module and C2f is used to enhance the extraction ability of spatial relationships and improve the ability to characterize features

    4、GB-Concat-在Concat模块中引入GLSA机制(由全局空洞自注意力(GASA)和局部窗口自注意力(LWSA)机制组成)自适应地将需要使用Concat的特征图进行上下文的整合,同时使用BiFPN,一种多尺度加权特征融合机制,学习不同输入特征的重要性

    GB-Concat - Introduce the GLSA mechanism in the Concat module (composed of the global empty space self-attention (GASA) and local window self-attention (LWSA) mechanism) to adaptively integrate the feature maps that need to be used in Concat for contextualization, and use BiFPN, a multi-scale weighted feature fusion mechanism, to learn the importance of different input features

    5、A-SPPF-在SPPF空间金字塔池化中加入add操作,促进梯度流动

    A-SPPF - A-SPPF-Add an add operation to SPPF spatial pyramid pooling to promote gradient flow

✍️ Recently

近日更新说明:

将个人改进项目模块整体打包,同时附带所需注意力模块(部分可酌情使用),baseline为YOLOv8,使用ultralytics8.2.0架构,部分结构的yaml已被删除(暂时不能公开),具体改进可参考./cfg/models/v8路径下的部分改进yaml,训练源码已提供(train.py)。项目同时提供了FPS计算(已整合至训练中,也可使用FPS.py单独计算)、基本的微调训练、剪枝代码(感兴趣的可以参考代码进行进一步设计,也可使用Torch-Pruning库对代码进行改进,部分剪枝模块不公开。

## ✍️ Authors - [@GuGuGuGun](https://github.com/GuGuGuGun) - Idea & Initial work ---

遵循MIT开源协议

License:MIT

About

自用改进模块-Self-Use Improvement Module

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published