Use Baidu PaddlePaddle easydl to complete garbage classification

One 、EasyDL Product introduction
EasyDL from 2017 year 11 From the middle of the month , Take the lead in launching targeted AI Zero algorithm basis or zero threshold for enterprise users pursuing efficient development AI Development platform , Provide data collection from 、 mark 、 Cleaning to model training 、 One stop shop for deployment AI Development capability . Customized for all walks of life AI For enterprise users who need , Whether you have AI Basics ,EasyDL Simple design , Easy to understand , The fastest 5 You can learn it in minutes ,15 Minutes to complete model training .
Collected original pictures 、 Text 、 Audio 、 video 、OCR、 Tables and other data , after EasyDL machining 、 Study 、 After deployment , Through public cloud API call , Or deployed on a local server 、 Small equipment 、 On the special adaptation hardware of the software hardware integration scheme , Through offline SDK Or private API Further integration , The process is as follows ：

Two 、 Dataset creation
To complete the garbage classification model training , First you need to prepare the dataset , Import data , Data analysis , Train again , After the model is obtained, it can be verified .
Use EasyDL front , First, you have to create a baidu account , After completing the real name authentication, proceed to the following steps .

（1） Click use now

​

EasyDL Official website address ：https://ai.baidu.com/easydl/

（2） Select image classification
Garbage classification data experience address ：https://aistudio.baidu.com/aistudio/datasetdetail/108025
Other types of data sets can also be searched on this website ： https://aistudio.baidu.com/aistudio/datasetoverview

（3） Create a dataset

​

​

Set dataset name , Set to garbage classification . The dataset has been classified in different folders , It belongs to marked information , Classify by folder name .

​

​

（4） Data analysis

​

​

3、 ... and 、 Training models
After the dataset annotation is completed , Next, create the model , Start training .
（1） Click create model

​

（2） Fill in information

​

（3） Choose training

​

（4） Configure training parameters
EasyDL Support multiple deployment methods , You can choose according to the environment you use .
I need to use my local training equipment here , Here I choose to deploy locally , Choose a universal small device . Select the data set just marked .

​

Then start training . At present, there is free computing power , You can also spend money on higher computing power , The training speed will be much faster , Of course , Free is generally enough , Time is also fast .

​

Then wait for training , When the training is finished, email 、 You'll get a reminder by SMS .
Put the mouse cursor here , You can see the progress of the training .

​

（5） Training done
After a period of waiting , The model has been trained .
Through the training results, we can see , My model accuracy is 93.66%, The result of the training is very good .

​

Four 、 Publish model
After training , Then release the model , After the model is released, it can be downloaded and used .

​

My environment here needs to be used locally and offline , Here it will be published as a browser / Applet .

​

The released platform supports a variety of systems ,Windows,MacOS, Android.

​

Then it was released , Wait for a while , After publishing, you can download .
5、 ... and 、 Download the model for testing
（1） Applet experience
Select the left option bar browser / Applet deployment —> browser / The applet service finds the list of published models , Download the corresponding model . I'll download the most accurate acceleration model here .

​

Click on the applet experience , Use baidu APP Scan QR code and upload pictures to quickly experience .

​

Applet results

​

（2） Browser deployment experience

​

Here we are , The whole training process has been completed .
6、 ... and 、 summary
Through this tutorial , You can see EasyDL It is very convenient to use , From the analysis of data 、 Model training to the final model deployment , One stop solution AI All the problems of application . Even Xiaobai users can easily complete model training to deployment , And the iteration speed is very fast . Last , I hope everyone is in AI Use... On applications EasyDL Develop more functions , come on. ！

This article is shared in Blog “ Flying propeller PaddlePaddle”（CSDN）.
If there is any infringement , Please contact the [email protected] Delete .
Participation of this paper “OSC Source creation plan ”, You are welcome to join us , share .