Landmark recognition models

Within AITrack there are several neural networks trained to recognize 68 facial landmarks which are later used for estimating user's head position. These networks are named as follows:

• Fast: This is the fastest model, which should have the less CPU footprint on your machine, but it's also the least accurate. Its usage is only recommended on cases where there aren't too many resources available or when the other models worsen your gameplay experience.

• Medium: Offers a good tradeoff between speed and accuracy. It's the recommended baseline for most users.

• Heavy: The most accurate model, but also the most CPU intensive. Use it when your hardware allows it.

• Very-Fast (since v0.7.0): This is an EXPERIMENTAL model which should run a little bit more faster than "Fast" but being closer in accuracy to "Medium". Please, note that, as this is experimental, could change in future versions.

As a final remark, all models should work well under good lightning conditions, so make sure your face is well lit when using AITrack.