当前位置:网站首页>More, faster, better and cheaper. Here comes the fastdeploy beta of the low threshold AI deployment tool!
More, faster, better and cheaper. Here comes the fastdeploy beta of the low threshold AI deployment tool!
2022-06-28 03:09:00 【Paddlepaddle】
This article has been published on the official account of the flying oar , Please check the link ;
Achieve greater, faster, better and more economical results , Low threshold AI Deployment tools FastDeploy Here comes the beta
The application of artificial intelligence industry is developing faster and faster , The adaptation deployment that developers need to face is becoming more and more complex . One side , New algorithm models emerge in endlessly , On the other hand , Various AI Hardware chips are also emerging . meanwhile , In the actual algorithm implementation , It also needs to deal with different scenarios ( Server deployment 、 Service deployment 、 Embedded deployment 、 Mobile terminal deployment, etc ), Different operating systems (Linux、Windows、Android、iOS etc. ), Different programming languages (Python、C++ etc. ), These are AI The implementation of the developer project brings no small challenge .

For the above pain points , We began to try a new product FastDeploy, It can aim at the important in the industrial landing scenario AI Model , In different hardware environments , Support developers to download a variety of precompiled SDK, Developer reference documentation , It can be done in a few simple steps AI Model deployment , Greatly reduce deployment difficulty and time cost .
Now? ,FastDeploy Released 0.1 edition , This version has three types of features :
Low threshold : One line command download SDK, One line of command to experience deployment Demo; Consistent code experience on different hardware platforms , Reduce the difficulty of secondary development based on business model and business logic ; Officially verified , The model accuracy and reasoning robustness of the algorithm are guaranteed .
Multi scene : Support the cloud 、 edge 、 End ( Including mobile end ) Enrich the software and hardware environment deployment , Support image 、 Video streaming 、 One click Service and other deployment methods , Meet the needs of different scenarios AI Deployment appeal .
Multiple algorithms : Support image classification 、 object detection 、 Image segmentation 、 Face detection 、 Detection of key points of human body 、 Text recognition, etc 30 Model selection of several mainstream algorithms .
Click to get the project portal
https://github.com/PaddlePaddle/FastDeploy
Low threshold use
One line command to quickly experience deployment
FastDeploy By way of configuration file , Realization AI Pre and post processing of the algorithm , Novice developers can also easily deploy . Different platform API The interface definition style is unified , It is convenient for developers to carry out flexible secondary development based on business model and business logic . For deployment and compilation challenges in the development environment ,FastDeploy It provides developers with an officially validated compilation environment and compilation scripts , Reduce AI The threshold for deployment and landing . at present FastDeploy Two types are available SDK Download mode :
- Page experience , Available for download .

- GitHub Command line experience , With a simple line of command , You can download the corresponding SDK. Developers can refer to the documentation , Quickly download the SDK Deploy to the actual project .


Multi scene coverage
Multiple platforms at the edge of the cloud can be caught in one net
FastDeploy v0.1 At present, we support Intel X86_64 CPU、NVIDIA GPU X86、ARM CPU(v7、v8)、Jetson Nano、Jetson TX2、Jetson Xavier Series hardware , And support Windows、Linux、Android、iOS Four operating systems ; Support image 、 Video streaming ( Include RTSP、RTMP、HLS、HTTP And other common network protocols ) Input reasoning deployment , And it can be deployed and experienced as a service ; In most scenarios, it supports C++ Deployment and Python Deploy , It meets the different deployment demands of different developers .

Multi algorithm support
Enabling different industries AI to ground
FastDeploy v0.1 Cover 7 Large class algorithms ( Image classification 、 object detection 、 Semantic segmentation 、 Instance segmentation 、 Face detection 、 Detection of key points of human body 、 Text recognition ), Provide access to 300 individual SDK. meanwhile ,FastDeploy It can quickly realize personalized model replacement , The developer only needs to replace the trained model with the deployment SDK in , You can complete the rapid deployment of your own business model .




Get started quickly
Developers only need to download the corresponding model 、 Hardware 、 Operating system SDK, One line of code to experience FastDeploy Rich deployment capabilities .

We use YOLOv5 For example , Take a look FastDeploy On different platforms (Linux System 、Windows System 、Android System 、iOS System ) Of SDK Engineering examples .
- YOLOv5 stay NVIDIA GPU / X86 CPU Windows Engineering examples under :

- YOLOv5 stay NVIDIA GPU / X86 CPU Linux Engineering examples under ( Include C++ and python Language ):

- YOLOv5 stay ARM CPU Android Engineering examples under :

- YOLOv5 stay ARM CPU iOS Engineering examples under :

FastDeploy There are two ways to use , Graphical interface mode and command line mode . The detailed operation steps are shown in the figure below .
- Graphical interface mode :

- Command line mode :

Learn more about
Want to learn more about how to use FastDeploy Fast reasoning , Sure :
- Refer to the corresponding documentation :

Slide down to see everything
- more FastDeploy Product information , You can see GitHub Learn more about :
https://github.com/PaddlePaddle/FastDeploy
- More participation is welcome AI Developers of model deployment , Join in FastDeploy Community communication .
Private online communication meeting
In order to fully communicate with developers about the problems encountered in the use of the product ,6 month 30 Friday night 20:30,FastDeploy The core members of the production research team will hold an online Tencent conference with the group developers , Discuss the current problems and future needs , Welcome to scan the code into the group , obtain 6 month 30 Daily online meeting link ! limit 200 mat , Until the report is full .
Official account of flying propeller , Sign up for an online exchange
Focus on 【 Flying propeller PaddlePaddle】 official account
Get more technical content ~
边栏推荐
- 树莓派-环境设置和交叉编译
- The first in the industry! MOS sub evaluation model for subjective video quality experience that can run on mobile devices!
- 横向滚动的RecycleView一屏显示五个半,低于五个平均分布
- 3年功能测试拿8K,被刚来的测试员反超,其实你在假装努力
- [today in history] June 1: Napster was founded; MS-DOS original author was born; Google sells Google SketchUp
- 英特尔锐炫A380显卡即将在中国面市
- [today in history] June 19: iPhone 3GS launched; Pascal was born; Anti terrorist elite begins testing
- "Everyday Mathematics" serial 53: February 21
- RichView TRVStyle
- [today in history] June 11: the co inventor of Monte Carlo method was born; Google launched Google Earth; Google acquires waze
猜你喜欢

Writing C program with GCC and makefile for the first time

Gateway microservice routing failed to load microservice static resources

Opencv -- Hough transform and some problems encountered

How to run unity webgl after packaging (Firefox configuration)

【活动早知道】LiveVideoStack近期活动一览

Le routage des microservices de la passerelle a échoué au chargement des ressources statiques des microservices

Gateway微服務路由使微服務靜態資源加載失敗

元宇宙标准论坛成立

音视频技术开发周刊 | 251

如何编写简洁代码?(上)
随机推荐
Severe Tire Damage:世界上第一个在互联网上直播的摇滚乐队
无心剑英汉双语诗004.《静心》
Online JSON to plaintext tool
读书,使人内心宁静
《天天数学》连载53:二月二十一日
The first place on the list - the carrying rate of front-end equipment is up to 10%, and the top 10 suppliers of digital key solutions
分布式事务TCC浅析
math_ (function & sequence) meaning of limit & misunderstanding and symbol sorting / neighborhood & de centring neighborhood & neighborhood radius
Publicity of the third batch of shortlisted enterprises! Annual Top100 smart network supplier selection
The first in the industry! MOS sub evaluation model for subjective video quality experience that can run on mobile devices!
Simple elk configuration to realize production level log collection and query practice
js清空对象和对象的值:
RichView TRVStyle ParaStyles
微信小程序中生成二维码
Why are so many people keen on big factories because of the great pressure and competition?
转载文章:数字经济催生强劲算力需求 英特尔发布多项创新技术挖掘算力潜能
Gateway微服務路由使微服務靜態資源加載失敗
面试:Bitmap像素内存分配在堆内存还是在native中
MFC CString to LPVOID
第一次使用gcc和makefile编写c程序