So I was exploring ways to deploy face landmark detector onto a mobile environment. The only way I know how was to either deploy with Tensorflow Lite or PyTorch Mobile.
Both of them are great in terms of inference speed and accuracy, but nothing beats the native SDK! Check out the Face API from Apple here
The advantages of ARKit over custom ML models here being:
- Ultra fast inference speed
- Short warmup time
- Buttery smooth animation
- Less than 50 lines of code! And no machine learning knowledge needed!
Disadvantages:
- Only supports devices from iPhone X and above
- It is not portable for Android counterparts. It means for business purposes, it is not possible to deliver the same experience for our customers using Android devices :(
Demo video :)