当前位置:网站首页>Use modelarts quickly, zero base white can also play AI!

Use modelarts quickly, zero base white can also play AI!

2020-11-06 20:10:19 itread01

Abstract : Don't miss it , see Copy How can siege lions use Huawei cloud ModelArts Play around AI.

“ since 2018 year 10 Since the release of the month ,ModelArts It has served hundreds of thousands of developers in many industries , Through the completeness of the basic platform, industry-oriented knowledge precipitation and platform ability , bring AI Application development is simpler and more efficient .”

Just as xuzhijun, the chairman in office of Huawei, is in 《ModelArts Artificial intelligence application development guide 》 It is mentioned in the preface of a book ,ModelArts Simplify AI Application development process 、 Optimized AI Application development costs 、 To reduce the AI Industry application development skills requirements . For a zero base partner , How to get started quickly AI Application development ? I believe you have read and I learned to use ModelArts The experience of , There should be an answer .

to encounter ModelArts

When it comes to bonding ModelArts, I have to tell you the story of Huawei cloud . Last year , Develop mobile application for Party A's father , Theirs. APP Distributed through Huawei application market , So I registered with Huawei developer account . At the same time , I am keen to participate in offline technology exchange activities in front-end development , I had the chance to get it by chance “HDC2019” Tickets for Huawei Developer Conference , The first focus is on front-end technology , For example, when I went to Songshan Lake in Dongguan, I was aiming at Huawei's fast application . stay HDC At the conference , What impresses me most is CodeLab, Similar to offline workshops , From the most basic “Hello World” Start , It makes it easy for developers to understand and accept a new technology . I remember that I was involved in several CodeLab, Basically, it's all about front-end development , Such as fast application development 、AppGallery Connect、 Mobile phone theme development, etc …… Of course , at present CodeLab Also a development guide on the wire , Such as :0 Code development image classification AI Model ; There are also more convenient sandbox Laboratories , For we can experience the use of ModelArts Realize flower image classification 、 Based on ModelArts To realize face recognition and other practical operations .

But after that , I've always focused on technology sharing in the front end , For example, Huawei cloud “7 R & D plays the front end ”…… Although the first contact with Huawei cloud did not directly encounter ModelArts, But for the later encounter did a lot of bedding . This year, 4 In the month , In the forum, I happened to pay attention to MDG Held 《 Innovation AI Activities : After reading the epidemic prevention class under cherry blossom in Wuhan University 》 And Mr. Mao Qichang 【AI Innovation developer Salon 】 How to use Huawei cloud ModelArts Realize strawberry ripening recognition ?, We have a preliminary understanding of Huawei's one-stop cloud service AI Development platform ModelArts, Zero basis can also develop such as image classification 、 Object detection 、 Predictive analysis 、 Sound classification 、 Text classification, etc AI Application development ; Then I also participated in Huawei cloud developers AI Youth class hacksong competition , It's the first time to adjust parameters , But it's zero basis after all , After a lot of trying , Finally, the baseline code was tuned to 0.948, And put this paragraph “ Keep hitting the wall ” Share the experience of ModelArts Section , So I was in ModelArts The first post of the section --《【 Touch and learn ModelArts】 Easy to finish Baseline And tune to 0.》, Later, I joined Huawei cloud AI Combat camp , I'm lucky to share my learning ModelArts In process “copy Experience ”, Completed the first live broadcast of life --【AI Actual classroom 】2020 Time is for AI Combat camp ——FasterRCNN and YoLoV3 Algorithm to complete object detection , Then it was really out of order , Deeply in love with ModelArts. She taught me :AI, It's not that hard ; Learn from AI Just arrive huaweicloud.ai!

Learning ModelArts

Before learning any knowledge or skill , I think we have to ask ourselves first , Why study ? For example, why to learn ModelArts? For me , The idea is simple : Now , Artificial intelligence seems to be integrated into daily life , There are only two kinds of people in the future -- One that makes money with artificial intelligence , One is for artificial intelligence consumption ; As the front end Copy Siege lions , If there can be AI Blessing automatic Copy Code is not like a fish in water , Isn't it fun ? There is no such thing as AI Mobile phones supported by chips can't be called “ Flagship aircraft ” The era of , No AI The siege lions with bonus skills feel a little out of date . however , When it comes to learning artificial intelligence , The first thing you may think of is that technology is too difficult 、 You can't learn . If , You have a one-stop experience AI Development platform ModelArts, You will be as surprised as I am : It turns out that AI Development can be as simple as that ?

So how to learn ModelArts Well ? According to my past experience , I feel that no matter what new knowledge or skills are learned , Official documents are the most original learning materials . For example , I want to learn the whole scene deep learning framework MindSpore, Then I might visit MindSpore The official website of --https://www.mindspore.cn/, We can easily find documents 、 course 、 Code even hand - held teaching video . Not only that , We can also easily find ways to join the community and find good friends who share the same ideals and move forward together .

Empathy , To learn ModelArts, We need to know her first step , First of all, the official website went on a wave :ModelArts It's built by Huawei cloud , Visit Huawei cloud ModelArts Zone and Huawei cloud AI Communities can easily find ModelArts And AI Related tutorials . such as , We can also get AI Developer growth path , From 0 To 1 Become AI Development talent .

In addition to official documents and tutorials , We can focus on MDG(ModelArts The developer community )、 Hua Weiyun ModelArts Forum and blog , The latest first-hand information can be obtained in time , Such as ModelArts New features 、 The work released by the government and the wonderful sharing by the community partners . secondly , Learn to ModelArts The best way to do it yourself is to do it yourself , Early learning , We can learn some basic concepts and operations according to the official documents , Then we can try to practice our own ideas , For example, Mr. Zhu Yongchun shared many creative practices before , There is milk tea identification 、 There is a fancy way to express 、 There are cloud graduation photos generated . Of course , Sharing is also helpful to your own promotion , Sharing is also a catalyst for learning , Looking forward to seeing your sharing 、 Learn together, make progress together !

Hello,ModelArts

Not much nonsense , First of all “Hello ModelArts” Case study -- Look for yunbao ( Using automatic learning to achieve object detection applications , Yunbao is the mascot of Huawei cloud ), I hope to do some short exercises by myself , You can feel the use of ModelArts Develop AI How simple the application is .

0. Environmental preparation

For the first time ModelArts Before , It is necessary to register Huawei cloud account and conduct real name authentication ;ModelArts When serving , You need to configure the global settings first , You can delegate ModelArts Visit OBS、SWR、IEF And rely on services , With delegation, you can control the scope of authorization more finely , Or by configuring the access key for authorization . This case also requires data storage services OBS, You can refer to ModelArts Preparation work .

1. Data set preparation

Benefit from ModelArts Powerful AI Market provided data set sharing function , I have uploaded the data set of this practice , Just visit ModelArts AI The market's dataset module , Search for “Yunbao”, You can find some of the marked yunbao image collection . Through AI Distribution of market data set modules , We can download the data set directly to our own ModelArts in . Data set management allows direct management of the slave AI Market downloads ModelArts Data sets , In this way, there is no need to download the dataset locally and then upload it to OBS We've created a new dataset , Greatly simplifies the development process .

2. Data label

Based on the data set downloaded in the previous step , We can create a new automatic learning - Object detection task , Mark the data in the task ; It can also be done in data management - Select the data set you downloaded in the previous step to label it . Of course , If you want to experience automatic grouping 、 Intelligent annotation and other data sets high-level operation , It is suggested that a second way should be used , After labeling, synchronize the data in the automatic learning task .

3. Training

As part of the information provided in this practice has been marked , So we can train directly , But the results of the training are not as good as all the marks . Building training is also very simple , Gently ,ModelArts And then we started to train automatically , Just wait a moment .

In the picture below V001 It's just marked 26 The training results of the pictures in ,V002 It's all marked training results . The difference is very obvious , Of course, the training time difference is quite big , All marked training hours are almost the same 2 times , But what comes in is a significant improvement in precision , It's also worth it .

4. Deploy

Deployment is just a click , I'm here to V002 This version is based on online deployment , Still waiting for a moment , Call when the service is running normally . I have to praise it here ,ModelArts It still provides a free online deployment service example , And further reduced our development AI Cost of .

Simple debugging , Basically, it can identify cloud treasure and its location accurately , So far zero code development object detection AI The model is done , you Get Have you arrived ? Is it right? So easy?

Explore happiness , Share happiness

Just learning ModelArts When , There's a little thought : I organized my learning experience into documents or videos to share, helping more partners to understand and learn ModelArts, Even published a comprehensive study ModelArts E-book tutorial for …… However , A few months passed , My thoughts are just ideas , The official development guide has indeed been published ! But , The idea is still in my mind , More detailed tutorials still need to be polished , There are still small goals -- In the next year, learn and sort out 《 Touch and learn ModelArts》 E-books , First blow out the cowhide .

Looking back on the past few months of study , except AI The combat camp insists on clocking in and sharing , Based on ModelArts And the bank's front-end development , I had a one-stop experience AI Develop : Based on ModelArts Official case completed bank deposit forecast small Demo, See 【 Touch and learn ModelArts】 Did you make a deposit today ?; Based on AI The named entity recognition case of the combat camp completes Huawei cloud application magic cube AppCube And ModelArts Combined exploration , See 【Copy Siege lion log 】ModelArts And AppCube Double “ magic ” It's a double festival . Although there is no technical content , It's not a climate , But it's also step by step , Study and explore in a down-to-earth manner , At the same time, thank you for your love , It feels very worthwhile , Very happy .

besides , I'm also keen on giving ModelArts File error correction for sample code library , It's real “ File modification contributors ”; Of course, through learning ModelArts-Labs Code library , It is also the first time to master ModelArts New play ; If you can go beyond learning , And by the way, help other kids , Why not ?

The link in the article is

  • [1] Sandbox lab : https://lab.huaweicloud.com/
  • [2] Use ModelArts Realize flower image classification : https://lab.huaweicloud.com/testdetail.html?testId=287
  • [3] Based on ModelArts Realize face recognition : https://lab.huaweicloud.com/testdetail.html?testId=337
  • [4]《 Innovation AI Activities : After reading the epidemic prevention class under cherry blossom in Wuhan University 》: https://bbs.huaweicloud.com/forum/forum.php?mod=viewthread&tid=49066
  • [5]【AI Innovation developer Salon 】 How to use Huawei cloud ModelArts Realize strawberry ripening recognition ?: https://bbs.huaweicloud.com/forum/forum.php?mod=viewthread&tid=51063
  • [6] Huawei cloud developers AI Youth class hacksong competition : https://competition.huaweicloud.com/information/1000040170/introduction?track=111
  • [7]《【 Touch and learn ModelArts】 Easy to finish Baseline And tune to 0.》: https://bbs.huaweicloud.com/forum/thread-53839-1-1.html
  • [8]【AI Actual classroom 】2020 Time is for AI Combat camp ——FasterRCNN and YoLoV3 Algorithm to complete object detection : https://bbs.huaweicloud.com/videos/102783
  • [9] MindSpore: https://www.mindspore.cn/
  • [10] Modelarts: https://www.huaweicloud.com/product/modelarts.html
  • [11] Hua Weiyun AI community : http://huaweicloud.ai/
  • [12] Milk tea recognition : https://bbs.huaweicloud.com/blogs/163274
  • [13] A fancy confession : https://bbs.huaweicloud.com/blogs/159114
  • [14] Cloud graduation photo generation : https://bbs.huaweicloud.com/blogs/174983
  • [15]【 Touch and learn ModelArts】 Did you make a deposit today ?: https://bbs.huaweicloud.com/forum/thread-63090-1-1.html
  • [16]【Copy Siege lion log 】ModelArts And AppCube Double “ magic ” It's a double festival : https://bbs.huaweicloud.com/blogs/198313
  • [17] ModelArts-Labs: https://github.com/huaweicloud/ModelArts-Lab

 

Click to focus on , The first time to learn about Huawei's new cloud technology

版权声明
本文为[itread01]所创,转载请带上原文链接,感谢