takshingc / azure-customvision Goto Github PK

The Custom Vision Service is an Azure Cognitive Service that lets you build custom image classifiers. It makes it easy and fast to build, deploy, and improve an image classifier. The Custom Vision Service provides a REST API and a web interface to upload your images and train the classifier. The Custom Vision Service works best when the item you're trying to classify is prominent in your image. Few images are required to create a classifier or detector. 50 images per class are enough to start your prototype. The methods Custom Vision Service uses are robust to differences, which allows you to start prototyping with so little data. This means Custom Vision Service is not well suited to scenarios where you want to detect subtle differences. For example, minor cracks or dents in quality assurance scenarios.

You must have the following to complete this lab:

• Azure Subscription
• Windows 10
• Microsoft Visual Code (latest update)
• PIP library
  • Pip install opencv-python
  • Pip install requests
  • Pip install azure-cognitiveservices-vision-customvision
  • Pip install matplotlib

This Hands-on lab includes the following exercises:

1. Create Azure Service
2. Build image classify project by custom vision service
3. Build image classify project by Python with visual studio code
4. Build image object detection project by custom vision service

Estimated time to complete this lab: 90 minutes.

1. Login Azure portal with provide Azure subscription

2. Create a Custom Vision Service

   • Create a new Resource Group, click Add at top panel. (remember Resource Group Name, please create all of services with this Resource Group)

   • Entry custom vision in Filter, you will find Custom Vision (Preview) in the list, choose this service and click create.

     

   • A new blade will open with information about the Custom Vision (Preview).

     

   • In the Custom Vision (Preview) blade, provide the requested information about your service as specified in the table below image.

   • Entry Name of your service select default resource group in Resource group section.

   • Click Create

1. Download and unzip customvision.zip to C drive from https://iothubstorageaccts.blob.core.windows.net/cvpic/customvision.zip

2. Open https://customvision.ai/ by Microsoft Edge

3. Click sign in with provided azure subscription account. You will see below page once login.

   

4. Click NEW PROJECT provide the requested information about your service as specified in the table below image.

   

   Entry Name of your project select resource group you create in Exercise 1 in Resource Group section, other by default.

   Click Create project

5. Click Add images on the top of panel, locate to folder C;\Lab\customvision\gear_images\axes, select all of images by Ctrl + A, click Open, entry axes as tag at My Tag section

   

   Click Upload 79 files

   Do the same action for rest tags in C:\ gear_images\

   

6. Click Train at left of top panel to train the model by provide images of each tag

   

   Waiting for training integration complete

   

   You will find Precision and Recall number of your trained model. Click Mark Default at top of panel

7. Click Quick Test at left of top panel, entry

   https://www.alpinetrek.co.uk/1500_1500_90/002-0808/berghaus-baffin-island-shell-jacket-hardshell-jacket.jpg

   In Image URL section, click you will see test result like below

   

   Close output panel.

8. Verify API by post URL endpoint. Click Predictions at top of panel, select result picture you just test, and click delete. Click View Endpoint

   

   Open Postman on the desktop, entry POST URL with first URL In above output panel. Entry header by key/value above

   

   Entry Body with raw and JSON format with

   {"Url": "https://www.alpinetrek.co.uk/1500_1500_90/002-0808/berghaus-baffin-island-shell-jacket-hardshell-jacket.jpg"}

   

   Click Send, you will see test result like below

   

1. Open Visual Studio Code on desktop, select Python 3.5 at left of bottom panel.

   

   clicK File at left of top panel, click Open File…, locate to C;\Lab\customvision\classify.py

   click Open

   

2. Open portal https://customvision.ai/, click sign in if your session is expired. Click setting icon at left of top panel,

   

   Copy Training Key and Prediction Key under resource group you created in Exercise 1 (e.g. techsubmit not Limited trial). Replace training_key and prediction_key in classify.py

   

   Replace project name classifylabfreeze with a new project name, different from project created in Exercise 2

   

   Replace PredictionEndpoint with Prediction Key above

   

   Click F5, select Python at output command

   

   Waiting for process complete, you will see the whole project was created, image uploaded, trained and verified automatically. (you could open https://customvision.ai to check project status, e.g. how many images uploaded)

   

1. Open https://customvision.ai/ by Microsoft Edge

2. Click sign in with provided azure subscription account. You will see below page once login.

   

3. Click NEW PROJECT provide the requested information about your service as specified in the table below image.

   

   Entry Name of your project select resource group you create in Exercise 1 in Resource Group section, select Object Detection (preview) in Project Type section.

   Click Create project

4. Click Add images on the top of panel, locate to folder C;\Lab\customvision\gear_images_detection, select all of images by Ctril + A, click Open

   

   Click Upload 251 files

5. After uploaded all files, click first picture, select boundary of helmets in the picture and add tag helmets like below.

   

   Entry helmets in Add Region Tag section first time.

   

   Click next icon and do same action of rest of all pictures, it might take couple minutes. (you don’t need tag all of pictures, but at least 30 tagged pictures need)

6. Click Train at left of top panel to train model by provide images of each tag.

   

7. You will find Precision, Recall and M.A.P number of your trained model. Click Mark Default at top of panel

8. Click Quick Test at left of top panel, entry. In Image URL section, click -> you will see test result like below

   

9. Verify API by post URL endpoint.

   Click Predictions at top of panel, select result picture you just test, and click delete.

   Click View Endpoint

   

   Open Postman on the desktop, entry POST URL with first URL In above output panel. Entry header by key/value above

   

   Entry Body with raw and JSON format with

   {"Url": "https://upload.wikimedia.org/wikipedia/commons/thumb/b/b6/Cologne_Germany_Industrial-work-with-Personal-Protective-Equipment-04.jpg/1200px-Cologne_Germany_Industrial-work-with-Personal-Protective-Equipment-04.jpg"}

   

   Click Send, you will see test result like below