Introduction
At Build 2017, Microsoft demonstrated the Custom Vision Service. This extremely cool service enables you to identify your own objects and things in images. It is easy, just upload a set of 6 to 12 images, hit the train button, and start calling the REST API.
The Custom Vision Service is part of the larger Azure Cognitive Services collection. The best use case for this tool is if you have a specific collection of unique things. If you just want to detect common things in images you're better off using the Computer Vision API.
Step 1 – Sign in
Navigate to - customvision.ai. Next sign in
Step 2 - Create a new project
In this step, I will create a new project to train a model.
This model should recognize the Eiffel tower and the triumphal arch in Paris. After clicking on create a new project, give a name, and choose the correct domain for this project.
Step 3
Upload your images (Eiffel tower and triumphal arch), first the Eiffel tower, and secondly the triumphal arch. Doing it this way will train the model to take pictures from all different angles.
After uploading give the images some tags.
Next, add the second set of images,
Step 4 - Training the model
Click on the green button. This step takes a few minutes. After training, you obtain two graphs that show information about your model.
The precision parameter tells you:
How likely is your classifier to correctly classify the image?
·
How likely is your classifier out of all images used to train the classifier?
Recall that this parameter tells you:
Out of all images that should have been classified correctly, how many did your classifier identify correctly?
Step 5 - The validation of the model.
You can test to validate if the model is working, by adding a new image
Step 6 - Connecting Prediction API
Our model is trained now, so we can use it in an application by publishing an API endpoint. To connect to your model, you must use the API prediction, so you must specify the prediction key in the section of the header and the section of iterationId.