当前位置:网站首页>yolov7 innovation point
yolov7 innovation point
2022-08-02 09:57:00 【ffllxx123】
Innovation 1: Extended Efficient Layer Aggregation Network E-ELAN and Composite Model Reduction
In the middle of the picture, there is actually a convolution and a siluA bn layer.64 on the right is the number of output channels, 1 is the size of the convolution kernel, and 1 is the stride.
v7 proposes e-elan, in fact, e-elan istwo elan.
Model scaling
From (a) to (b), we observeAs a result, when performing depth scaling on cascade-based models, the output width of the computational block also increases.This phenomenon will cause the input width of subsequent transport layers to increase.Therefore, we propose (c) that when performing model scaling on a cascade-based model, only the depth in the computation block needs to be scaled, and the rest of the transport layer uses the corresponding width scaling.You can make the width unchanged (roughly this reason).
Reparameterized Network
As you can see from the leftmost figure, the residualThere are two types of structures, the middle one is the re-parameterized residual structure. It can be seen that each residual structure only spans a 3x3 convolution, and sometimes the two residual structures are used together, but there is no inference stage.Residual structure, which makes the training accuracy higher and the inference speed faster.
Parameter fusion, that is, 3x3 convolution kernel 1x1 volumeProduct and do nothing These 3 can be fused into a 3X3 convolution.
both left and right are doing nothing, is the identity all the way.Then, for example, the original road has an ordinary 3x3 convolution, then adding this ordinary 3x3 convolution and the convolution on the right side of the above picture one by one can achieve parameter fusion.You get the effect on the far right of the image below.Re-parameterization is achieved.
This repconv is a reparameterized convolution.The author found that the reparameterized convolution plus the residual is not good (d figure),
Innovation point 3 tag matching
a picture is a common pyramid model, introduced in v7The b structure is added and the auxiliary head is added. You can see it in the c picture, and calculate the loss of the guide head and the auxiliary head at the same time. As you can see in the d picture, the distributor in the guide head will assist in calculating the loss of the auxiliary head, and then at the same time for the guide head.The loss and the loss of the auxiliary head are optimized by gradient descent.
边栏推荐
- 理解JS的三座大山
- QT专题:事件机制event基础篇
- Rust 从入门到精通03-helloworld
- Getting Started with SCM from Scratch (1): Summary of Background Knowledge
- 练习-17
- 【云原生】快出数量级的性能是怎样炼成的?就提升了亿点点
- The R language uses the rollapply function in the zoo package to apply the specified function to the time series in a rolling manner and the window moves, and set the align parameter to specify that t
- 百战RHCE(第四十六战:运维工程师必会技-Ansible学习1-基础知识讲解)
- The perceptron perceptron of Li Hang's "Statistical Learning Methods" notes
- 使用scrapy 把爬到的数据保存到mysql 防止重复
猜你喜欢
iNFTnews | 看见元宇宙的两面,何谓全真互联网和价值互联网?
Nodejs3day(express简介,express创建基本Web服务器,托管静态资源,nodemon下载及出现的问题,中间件,编写GET,POST,JSONP接口)
DVWA Clearance Log 2 - Command Injection
后管实现面包屑功能
软件测试H模型
新“内卷”席卷科技圈,Google CEO 要求 174000 员工提高工作效率!
【New Edition】DeepFakes: Creation, Detection and Influence
AutoJs学习-实现科赫雪花
HikariCP database connection pool, too fast!
高效时代,电商运营如何靠RPA快速提效?
随机推荐
matlab-day02
百战RHCE(第四十七战:运维工程师必会技-Ansible学习2-Ansible安装配置练习环境)
记某社区问答
后管实现面包屑功能
读博一年后对机器学习工程的思考
The 17th day of the special assault version of the sword offer
Redis数据结构
R语言使用zoo包中的rollapply函数以滚动的方式、窗口移动的方式将指定函数应用于时间序列、设置align参数指定结果数据中的时间标签取自窗口中的位置(参数right指定取自窗口的最右侧)
牛客网项目2.7开发注册功能 报错This application has no explicit mapping for /error......
system_error错误处理库学习
Spend 2 hours a day to make up for Tencent T8, play 688 pages of SSM framework and Redis, and successfully land on Meituan
Have you ever learned about these architecture designs and architecture knowledge systems?(Architecture book recommendation)
要长续航还是更安全?海豹与深蓝SL03对比导购
软件测试H模型
【OpenCV】-霍夫变换
mysql连接池的实现
AutoJs学习-AES加解密
iNFTnews | Seeing the two sides of the metaverse, what is the true Internet and the Internet of value?
【新版干货书】深度伪造 (DeepFakes):创造,检测和影响
李航《统计学习方法》笔记之k近邻法