当前位置:网站首页>Bisenet features
Bisenet features
2022-07-07 08:15:00 【I am a little rice】
1. Dual branch network
A network emphasizes details , Its characteristic graph is at least as small as 1/8 The size of the , In other words, try to keep more details of the feature map ;
The second is lightweight , He only uses three convolution layers , And that's what happened
The third is the branch of high-level semantic features (context path), Emphasize the deeper depth , The deepest can reach 1:32 Under sampling ,
The fourth is the lightweight model , Use inception To achieve , While maintaining a relatively deep depth . Still keep a relatively small amount of calculation , Through global average pooling , Realize the capture of context information , The main idea is simple and lightweight backbone network
The fifth is feature fusion , The characteristic graph space of the two branches is different , There will be great differences in characteristics , Fusion will lead to many noise points , So it uses batch normalization , Reduce differences between feature representations , Weighted fusion using attention mechanism , To emphasize high-level semantic features
2. Feature fusion module
First, stack on a channel , That is, in series , Then a common convolution is used for processing after the channel is stacked , Then a mechanism similar to attention is used , Finally, there is a short circuit
3. Loss function
Use the main loss function to monitor the whole network , Use the auxiliary loss function to monitor the deep network on the right , The main purpose is to speed up the training progress
边栏推荐
- JS cross browser parsing XML application
- Niu Mei's mathematical problem --- combinatorial number
- Interactive book delivery - signed version of Oracle DBA work notes
- 太真实了,原来自己一直没有富裕起来是有原因的
- The largest 3 same digits in the string of leetcode simple question
- 利用 Helm 在各类 Kubernetes 中安装 Rainbond
- Find the mode in the binary search tree (use medium order traversal as an ordered array)
- Network learning (III) -- highly concurrent socket programming (epoll)
- Merging binary trees by recursion
- JS quick start (I)
猜你喜欢
央视太暖心了,手把手教你写HR最喜欢的简历
buureservewp(2)
Bugku CTF daily one question chessboard with only black chess
Qinglong panel - today's headlines
Rainbond 5.7.1 支持对接多家公有云和集群异常报警
CCTV is so warm-hearted that it teaches you to write HR's favorite resume hand in hand
在Rainbond中一键部署高可用 EMQX 集群
Uniapp mobile terminal forced update function
在Rainbond中实现数据库结构自动化升级
【踩坑系列】uniapp之h5 跨域的问题
随机推荐
发挥创客教育空间的广泛实用性
【雅思口语】安娜口语学习记录 Part2
Find the mode in the binary search tree (use medium order traversal as an ordered array)
Qinglong panel -- finishing usable scripts
Hisense TV starts the developer mode
Empire CMS collection Empire template program general
快解析内网穿透为文档加密行业保驾护航
Summary of redis functions
Bayes' law
一文了解如何源码编译Rainbond基础组件
uniapp 移动端强制更新功能
Leetcode 187 Repeated DNA sequence (2022.07.06)
让Livelink初始Pose与动捕演员一致
提高企业产品交付效率系列(1)—— 企业应用一键安装和升级
Qinglong panel -- Huahua reading
LeetCode中等题之我的日程安排表 I
Recursive construction of maximum binary tree
机器人教育在动手实践中的真理
ZCMU--1492: Problem D(C语言)
OpenVSCode云端IDE加入Rainbond一体化开发体系