I assume that the average of these facial expressions are the average of all parameters, like how vertically open vs horizontally open a mouth is. However, human expressions are mathematical in nature, so how useful would a true average for two facial expressions be? I feel like any significantly different object would always average out to the same lukewarm look. Like a happy and sad face or a happy and angry face, would average out to an unreadable person.
Mehvix
@aidangarde If you're getting "average" results, you can always weigh them different proportions. Also, you could over/inverse exaggerate certain face shapes (i.e. using βi>1 or βi<0).
theflyingpie
Would similar problems to deformation in skinned meshes occur when interpolating between blend shapes? For instance, would eyelids sometimes clip through the eyeball, or lips clip through teeth if the interpolation between expressions is linear? If so, is there a way of systematically addressing these issues, or would they just have to be manually fixed?
zeddybot
I wonder if there is any equivalent for non-linear interpolation (e.g., using splines or Bezier curves) to blend shapes. If you want a face to move between a set of keyframe expressions at different time steps, you could use Bezier curves or Catmull-Rom splines to interpolate the positions of each vertex between the keyframe expressions.
I assume that the average of these facial expressions are the average of all parameters, like how vertically open vs horizontally open a mouth is. However, human expressions are mathematical in nature, so how useful would a true average for two facial expressions be? I feel like any significantly different object would always average out to the same lukewarm look. Like a happy and sad face or a happy and angry face, would average out to an unreadable person.
@aidangarde If you're getting "average" results, you can always weigh them different proportions. Also, you could over/inverse exaggerate certain face shapes (i.e. using βi>1 or βi<0).
Would similar problems to deformation in skinned meshes occur when interpolating between blend shapes? For instance, would eyelids sometimes clip through the eyeball, or lips clip through teeth if the interpolation between expressions is linear? If so, is there a way of systematically addressing these issues, or would they just have to be manually fixed?
I wonder if there is any equivalent for non-linear interpolation (e.g., using splines or Bezier curves) to blend shapes. If you want a face to move between a set of keyframe expressions at different time steps, you could use Bezier curves or Catmull-Rom splines to interpolate the positions of each vertex between the keyframe expressions.