you just can’t ignore >40% of the population
Anatomical variation is a surprisingly ignored consideration for face tracking/facial mocap in tech and entertainment. Simplified anatomy diagrams are often accepted as universally applicable to all faces and few further questions are asked.
The reality is: FACIAL MUSCLES ARE HIGHLY VARIABLE.
Some muscles required for standard facial action posing are:
- missing bilaterally
- missing unilaterally
- significantly variable in shape and/or size
These variations are not small. In fact, one of the muscles currently thought to be canon in the “basic emotions” is actually missing in up to 40% of the population!
If you are working on facial expression technology, it’s imperative to ask:
- Are we wasting valuable time and resources asking actors and research subjects to activate muscles they do not possess?
- Have we considered the impact that missing muscles can have on emotion and expression detection?
- Are we contributing to current issues with bias by not addressing these realities?
Understanding how we differ, where we differ, and how to take advantage of that information (rather than ignore it) will help us:
- reduce tracking bias
- increase data quality
- advance face tech more mindfully
Now for the good news / fun part . . .
what this has to do with dimples
A large percentage of people have muscle variations that create dimples when that particular muscle is activated. (NOTE: The dimples to which I am referring to are unrelated to the FACS “dimpler” – which is an action caused by the buccinator muscle.) Awareness of this information is extremely valuable to data labeling & facial rig creation.
Dimples are signifiers that can give you clues as to which part of the face is moving at a given time. If a labeler is unsure of an ambiguous action, knowledge of when certain muscle-based dimples are forming can give additional clues as to whether or not a particular facial action is currently taking place.
In the case of lip corner puller, many individuals (as many as 34% in one study) have a bifid zygomaticus major muscle – meaning, when they smile, the muscle used to raise their lip corners and widen their mouth is broken down into two insertion points. Understanding insertion points is critical for breaking down and recreating facial movements. If an individual has two insertion points, both need to be accounted for when representing that movement.
So far in my research, I have only looked into specific studies on the zygomaticus major muscle; however, these signals exist in other key player muscles as well and should be considered.
If anyone is interested in funding or inquiring about my research, please reach out at facetheFACS@melindaozel.com