The data for training face tracking algorithms generally consists of four components:
- Images: This component is a collection of images (still images or video frames) that contain an entire face. For best results, this collection should be specialized to the types of conditions (that is, identity, lighting, distance from camera, capturing device, among others) in which the tracker is later deployed. It is also crucial that the faces in the collection exhibit the range of head poses and facial expressions that the intended application expects.
- Annotations: This component has ordered hand-labeled locations in each image that correspond to every facial feature to be tracked. More facial features often lead to a more robust ...
