当前位置:网站首页>Popular science of data annotation: ten common image annotation methods
Popular science of data annotation: ten common image annotation methods
2022-06-24 11:44:00 【Lift up】
The rapid development of computer vision is inseparable from the support of a large number of image annotation data , With all kinds of image detection 、 Commercialization of recognition algorithm , The market is becoming more and more strict about the accuracy of image annotation , At the same time, for different application scenarios , Different image annotation methods are also derived .
1、 Semantic segmentation
Semantic segmentation is based on the attributes of objects , Area division of complex and irregular pictures , And mark the corresponding attributes , To help train the image recognition model , It is often used for automatic driving 、 human-computer interaction 、 Virtual reality and other fields .
2、 Rectangular box dimension
Rectangular box labeling is also called pull box labeling , It is the most widely used image annotation method at present , In a relatively simple way 、 Convenient way in image or video data , Quickly frame the specified target object .
3、 Polygon annotation
Polygonal annotation refers to in still pictures , Use polygon boxes , Mark irregular target objects , Dimension relative to rectangular box , Polygon annotation can frame the target more accurately , At the same time, for irregular objects , It is also more targeted .
4、 Key point marking
Key point annotation refers to the manual way , Mark the key points at the specified position , For example, facial feature points 、 Human bone connection points, etc , It is often used to train facial recognition models and statistical models .
5、 Point cloud annotation
Point cloud is an important expression of 3D data , Through sensors such as lidar , It can collect all kinds of obstacles and their position coordinates , The annotator needs to classify these dense point clouds , And mark different attributes , It is often used in the field of automatic driving .
6、3D Cube dimension
Different from point cloud annotation ,3D Cube annotation or annotation based on two-dimensional plane image , The announcer frames the edges of three-dimensional objects , And then get the vanishing point , Measure the relative distance between objects .
7、2D/3D Fusion annotation
2D/3D Fusion annotation refers to the simultaneous annotation of 2D and 3D Label the image data collected by the sensor , And build relationships . This method can mark the position and size of the object in plane and three-dimensional , Help the autopilot model enhance vision and radar perception .
8、 Target tracking
Target tracking refers to moving images , Frame extraction and annotation , Mark the target object in each frame , Then describe their trajectory , Such annotations are often used to train automatic driving models and video recognition models .
9、OCR Transcribe
OCR Transcribe is to mark and transcribe the text content in the image , Help train and improve the image and text recognition model . at present , Jing Lianwen supports simplified Chinese 、 Traditional Chinese 、 English 、 Japanese 、 Korean 、 French 、 German 、 Spanish 、 Transfer of printed or handwritten pictures in more than ten languages such as Arabic .
10、 Attribute discrimination
Attribute discrimination refers to the way of manual or machine cooperation , Identify the target object in the image , And mark it with the corresponding attribute .
边栏推荐
- 广发证券靠谱吗?开证券账户安全吗?
- 8 types of API automated testing that technicians must know
- Why does the virtual machine Ping the host but not the virtual machine
- [the lottery in May has ended, and the list of winners has been announced] special session of techo youth university open course database
- Programmers spend most of their time not writing code, but...
- Axi low power interface
- 12+!不同癌症中TMB与ICI反应之间的免疫相关因素研究
- Shell脚本(.sh文件)如何执行完毕之后不自动关闭、闪退?
- Insurance app aging service evaluation analysis 2022 issue 06
- Group policy export import
猜你喜欢
Moving Tencent to the cloud cured their technical anxiety
Linker --- linker
[Old Wei makes machines] issue 090: keyboard? host? Full function keyboard host!
How to write controller layer code gracefully?
How is the e-commerce red envelope realized? For interview (typical high concurrency)
Today in history: Turing's birth day; The birth of the founder of the Internet; Reddit goes online
GTest从入门到入门
Group counting_ Structure and workflow of CPU
PHP SMS notification + voice broadcast automatic double call
u盘安装kali并且持久化
随机推荐
Moving Tencent to the cloud cured their technical anxiety
Code is really - omnipotent! Refuse to fight
[Flink source code practice (I)] add a rest API to Flink
New progress in the construction of meituan's Flink based real-time data warehouse platform
Audio knowledge (III) -- MFCCs code implementation
【老卫搞机】090期:键盘?主机?全功能键盘主机!
Example of PHP observer mode [useful in the laravel framework]
Tools and methods - use code formatting tools in source insight
Insurance app aging service evaluation analysis 2022 issue 06
电商红包雨是如何实现的?拿去面试用(典型高并发)
《opencv学习笔记》-- 矩阵归一化 normalize()函数
u盘安装kali并且持久化
我在深圳,到哪里开户比较好?现在网上开户安全么?
"Meng Hua Lu" is about to have a grand finale. It's better to learn it first than to look ahead!
How to export only the titles in word documents? (i.e. delete all the text contents and keep only the title) stop B
I want 18K. Can I pass it?
如何开发短信通知和语音功能医院信息系统(HIS系统)
Cloud vendor secondary virtualization restrictions
[the lottery in May has ended, and the list of winners has been announced] special session of techo youth university open course database
Why choose b+ tree as storage engine index structure