Translated from 日本語 • View original This is Akiyama, an engineer.
In Japan, the second day of de:code is attracting attention to various services.
This time, we summarized the steps to try out the Custom Vision Service announced at Build2017 the other day.
Custom Vision Service Overview
Open the Custom Vision Service page directly or click Try Custom Vision Service from the Azure Custom Vision Service page .
Currently, there is no usage fee, perhaps because of the preview period.
Making a Classifier
Sign in to the Custom Vision Service.
Unlike the Azure portal, it is authenticated only for Microsoft accounts.
When you log in, a simple page opens with just a list of projects.
Create this classifier in New Project.
Next, register the image.
You need to register at least 5 tickets here. The number of patterns in the image leads to accuracy, so it's a good idea to have a variety of images.
Tag the image. This tag means that it is the training data.
Repeat the image registration as many tags as you want to determine to create the correct answer data.
Once you have enough data, start learning in Train.
This completes the AI just for you. Thank you for your hard work.
Verification
To try it out, upload an image from Quick Test.
You can see that the accuracy is coming out.
Afterword
Until now, the image recognition provided by Cognitive Services was limited to general-purpose applications, and if we were to work with industry-specific objects, would we implement machine learning ourselves? It was a situation. Custom Vision is just the right service to fill this gap. Let's take advantage of it.