Alibaba Cloud | Image classification by TensorFlow

Introduction

The Internet has a huge volume of data, some are structured, semi-structured and some are unstructured. The processing of the data requires the use of deep learning algorithms, moreover, the algorithms can even be daunting sometimes, requiring high-end GPU and great resources.

Well, Alibaba Cloud has made a way such that you could do your task effectively and efficiently without bothering about any such. This article introduces the method of image recognition using deep learning frameworks, which can be applied in scenarios as facial recognition, object detection, image filtering, etc.

DataSet

To download the dataset click here.

source - https://zos.alipayobjects.com/rmsportal/PrUynAHYaCQQLrneFpWN.png

Training Procedure

Create an experiment in the machine learning platform, you need to enable GPU usage and activate Object Storage service.

• Download dataset and decompress them
• Log on to OSS and create OSS bucket
• Create a new directory in the OSS bucket. An aohai_test directory is created.
  
  
  source - https://zos.alipayobjects.com/rmsportal/eXgLTWObHKpDvnWTWTVN.png

   

• Log into ML platform click settings and configure OSS permission
  
  
  source- http://docs-aliyun.cn-hangzhou.oss.aliyun-inc.com/assets/pic/75343/intl_en/1529915901755/fig03.png

Model Training

• Drag a Read OSS Bucket component and a tensor flow component to the canvas and configure Tensor Flow component as follows
  
  
  
  

   

source - http://docs-aliyun.cn-hangzhou.oss.aliyun-inc.com/assets/pic/75343/intl_en/1529916448860/fig29.png

• Click Run to start the training procedure, also you can change the GPU by changing the configuration.
  
  
  source - http://docs-aliyun.cn-hangzhou.oss.aliyun-inc.com/assets/pic/75343/intl_en/1529916499566/fig30.png

   

Result Prediction

Another TensorFlow component can be used in predicting.

source - http://docs-aliyun.cn-hangzhou.oss.aliyun-inc.com/assets/pic/75343/intl_en/1529917231397/fig31.png

The checkpoint picture used for predicting is:

source - https://zos.alipayobjects.com/rmsportal/ABdyULPNZjYkLGzzCuWZ.jpg

The predicted result is as follows 

source - https://zos.alipayobjects.com/rmsportal/NZHiAyfBLtOuYGbOozDO.png