Creating a Custom Inference API

(You can work in pairs, if you want to.)

Your task is to create an inference Web API that checks if an image contains a cat. This starting project already provides the ML functionality for this (src/CatPredictor.py), namely a class that uses two pretrained models from the ImageNet family (VGG16 and MobileNet). You can see an example of its usage in src/usage-example.py. An advanced skeleton for the API you should build is provided based on the Flask framework in src/prediction-api.py. You can use the HTTP client of your choice to test it, e.g., Postman or curl. Once your API is ready, encapsulate it into a Docker image, and build and run a container with it.

Bonus task if you are quick: change your client project from last tutorial to use your newly created API instead of the Google Cloud Vision API for identifying all cats from the image set.

Prerequisites

1. Install Python 3, preferably >= 3.9
2. Install dependencies: pip install -r requirements.txt
3. Install Docker Desktop (for the container part)

Usage

# using the prediction example (you can change the used filename to a different file under `images`)
python src/usage-example.py

# using the REST API
# set an ENV variable for the Flask entrypoint file
export FLASK_APP=src/prediction-api.py
# start the API --> API offers the endpoint `POST http://localhost:5000/cat-predictions`
flask run
# for enabling hot reload (execute before starting)
export FLASK_ENV=development