In homework I1 you will use third-party machine learning APIs and in the group project you will develop your own APIs. In this lab, you will experiment with both, connecting to the Azure Vision API and providing your own API endpoint. To receive credit for this lab, show your work to the TA during recitation.
- Create an account and connect to the Azure Vision API
- Explain to the TA why hard-coding credentials is a bad idea. Commit your code to GitHub without committing your credentials.
- Run the API endpoint with the starter code and demonstrate that it works with an example invocation (e.g., using curl).
Clone the starter code from this Git repository
The code implements a flask web application that receives API requests to analyze an image and return information about the image, including the text contained within. To identify the text, the OCR feature of the Azure Vision API [documentation, response format] can be used by adjusting the API endpoint and credentials in the code. We use the Azure’s provided libraries to abstract from low-level protocol details.
Install the dependencies in the requirements.txt file with pip or similar. To set up the flask server, just run python3 app.py. The system should try to analyze an example image and report the results when you go to http://localhost:3000/
-
Sign up for the a student account for Microsoft Azure: https://azure.microsoft.com/en-us/free/students/ – no credit card required
-
Create an instance of the Computer Vision service and get an API endpoint of your instance of the service.
-
Get a subscription key to authorize your script to call the Computer Vision API.
-
Update the code with the endpoint and key and test it.
The starter code hardcodes credentials in the code. This is a bad practice.
Research and discuss best practices, such as never hard-code credentials, never commit credentials to Git, rotate secrets regularly, encrypt your secrets at rest/in-transit if possible, practice least-access privilege on machines where your credentials are stored as environment variables or within local files.
Rewrite the code to load credentials from a file or an environment variable and commit the code without the credentials to GitHub.
The starter code comes with a flask server that serves the website at http://localhost:3000/ but also exposes an own API at http://localhost:3000/api/v1/analysis/ accepting a GET request with a JSON object with a single field “uri” pointing to an image to analyze.
Identify how to call your own API with a tool like curl or Postman.
Optionally extend the API or document it with Swagger.
- Redhat article on API
- Azure Computer Vision
- API Design Best Practices
- API Endpoint Best Practices
- The file seai-azure-cv-ocr-api.json has the structure to test calls to the Azure Vision API with Postman.