HowPretty

The most brutal app in human history. This app takes a picture of your face and scores it from 0 to 1 where 1 is the most attractive.

I'm not very pretty according to the app.

IMPORTANT NOTICE: THIS APP IS A JOKE AND SHOULD BE REGARDED AS SUCH. PRETTINESS IS SUBJECTIVE AND THE OUTPUT OF THIS APP SHOULD NOT BE TAKEN SERIOUSLY.

Building HowPretty

1. I started looking for a dataset which has images of faces along with labels for attractiveness. I used a subset of this dataset.
   • Many images.
   • Many labels.
   • Publicly available without registration.
2. I built and trained a model using TensorFlow 2 and Keras on Google Colab. You can view the model here. I also exported the model to a .tflite file which can be found here.
   • MobileNet for fast inference on mobile devices.
   • Used binary classification because there is only one class.
   • Trained the model only on input images and not other aspects because we don't have access to those at inference time.
3. I built a simple app which has a camera layer. See #b2c17cf.
   • Reused the code I've written for this tutorial with corresponding code. This code includes a camera layer and a shutter button.
   • Changed the code to use the FaceTime HD camera instead of the Camera.
4. I followed the TensorFlow lite iOS guide and added the model to my app.
   • I chose to use a local model to preserve the users privacy.

HowPretty works with the newest version of iOS 13.3.

Improvements

Currently HowPretty is an MVP. There are a couple of things that could be improved if this app goes into production:

• Add a nicer score UI. For example, add a label or "progress bar" for scores instead of an alert.
• Retrain the model on the full set of images instead of a subset.
• Add a loading screen.
• Design an app icon.
• Add an Android version.
• Use bounding boxes of face to crop the image for higher accuracy. Tell the user to move closer or further away to get the perfect resolution (150 by 150).
• Add a share buttons which shares the image, score and a link to the app.

---

©Rick Wierenga