Shulou(Shulou.com)06/02 Report--

Previously, we introduced the construction of Azure AI's face recognition service Demo, so how about object recognition? Today I will show you how to build a custom vision of Demo to achieve object identification. For example, I first use machine learning to train it to tell the cup what it looks like. Give him a picture of at least 5 cups to complete the training. Finally, I upload or take a picture or photo of any water cup, and it will tell me how much it looks like a cup. Of course, the more training materials, the higher the recognition accuracy and matching rate.

First create a custom visual service on Azure

Create

Intelligent Kiosk's custom visual Demo only supports the south-central part of the United States, so you can only choose the south-central part of the United States. F0 is free, but there is a limit on the number of calls, and the trial is enough.

After the creation, there are two cognitive services with custom vision, one is training, the other is prediction.

Click on the training custom visual cognitive service, and copy the key 1 in the training cognitive service key to the custom visual training key in the Intelligent Kiosk settings.

Then open the cognitive service of predicting custom vision, and copy the key 1 to the KEY of custom visual prediction in the Intelligent Kiosk setting.

Then switch back to Demo Gallery and select Custom Vision Explorer Demo

Built-in some trained items, such as basketball, blue box, athletes, here I click the + sign to train a group of subjects myself.

Click on the + sign for the new project

The name of the project is object, click add, and select general picture category

Create a new tag for mobile phone

Click on the + account on images and I upload 5 pictures of mobile phones that I find randomly on the Internet.

By the same token, create a new water cup tag and upload 5 randomly found water cup pictures on the Internet.

Upload 5 mouse pictures, and finally click on the training project in the upper right corner to let the machine learn to be familiar with what these objects are.

Finally, there is the test. I went back to select the target project that I just created. I clicked on the camera in the upper right corner, and I took a picture of the water cup in my hand. I could see that 85% of it looked like a water cup.

In the same way, I took another mouse and identified 71% like a mouse.

Take a picture of a mobile phone and identify it as a mobile phone (100% like)

Two objects are photographed in the same frame to identify the mobile phone and the mouse respectively.

You can also click on a cell phone photo on Search in the upper right corner to identify it.

The accuracy of recognition depends on the number of material libraries provided and the quality of the pictures. The more materials you train, the higher the accuracy and matching degree of the identified objects. This is to use Azure to customize the Demo of identifying objects. Of course, you can also train a lot of custom things, such as weather, vehicles, car brands, car parts and so on. As long as you upload the corresponding pictures and tag them. After training, any uploaded and photographed pictures or photos can be identified.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.