Inspiration

One of the authors always had to show his ID even though he's pretty old. As they say "Asians don't raisins."

On a more serious note, we realized Facial Recognition and Deep Learning complemented each other, so why not using them to create an age predictor.

What it does

You either upload a photo (for mobile and desktop users) or use your camera (only if you're on mobile) and the model would show its prediction of your age within a range.

How we built it

We used fastai and Pytorch to prepare, process, and clean the data, build the model, and fine-tune it. We used Voila and Binder to deploy it. And we built the model in Google Cloud Platform.

Challenges we ran into

Too large a dataset, cloud CPU was insufficient, an efficient way to extract the age of the person to validate the result, and deployment/

Accomplishments that we're proud of

We went through the whole process: data gathering, processing, building, cleaning, tuning, and deployment. The UI is intuitive, user-friendly, and to-the-point.

What we learned

The time it took for each step was 2-3 times more than expected, and the scale of the project was smaller than suggested.

Finished is more important than scale or perfection.

So from the beginning, both of us were willing to scale to get this done. And it paid off well.

What's next for AI Age Prediction

More data, better metrics, and more fine-tuning to reduce the confidence range.

Built With

  • azure
  • binder
  • fastai
  • gcp
  • voila
Share this project:

Updates