Predicting Geolocation from image using machine learning

This repo houses the code for predicting geolocation from a given image url in form of a restful API.

The model, pre-processing code and labels used in this project comes from Multimedia Berkeley Group the model is a MxNet model trained on YFCC100M Multimedia Commons dataset.

Inspired from @ranman's Twitch livestream AI Twitter bot session

This project was built to test the code locally, however if you follow the above interactive coding session you should be able to deploy this on AWS easily.

• Download the model using python download.py

• Execute docker-compose up

  • This will build the image based on the Dockerfile and start the API at http://localhost:8080

    • You can start testing the API using cURL or Postman

    curl -X POST \
    http://localhost:8080/invocations \
    -H 'Content-Type: application/json' \
    -d '{
        "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/a/a1/Statue_of_Liberty_7.jpg/250px-Statue_of_Liberty_7.jpg",
        "max_predictions": 5
        }'

  • You should see top 5 predictions returned in JSON array sorted by probability of the geolocation.

    [
        {
            "city": "New York City",
            "country": "US",
            "lat": "40.71427",
            "lon": "-74.00597",
            "region": "New York"
        },
        {
            "city": "New York City",
            "country": "US",
            "lat": "40.71427",
            "lon": "-74.00597",
            "region": "New York"
        },
        {
            "city": "New York City",
            "country": "US",
            "lat": "40.71427",
            "lon": "-74.00597",
            "region": "New York"
        },
        {
            "city": "New York City",
            "country": "US",
            "lat": "40.71427",
            "lon": "-74.00597",
            "region": "New York"
        },
        {
            "city": "New York City",
            "country": "US",
            "lat": "40.71427",
            "lon": "-74.00597",
            "region": "New York"
        }
    ]