|
In recent years, deep-learning method of facial landmark localization has been largely developed. The performances are continuously improved in facial landmark localization tasks. However, it also turns marking process into a black box and makes it unconvertable randomly in marking methods. In the end, there is little previous research interprets which landmarks to choose will affect better. The 68-landmark method is the most popular. In 3D reconstruction field, most studies use this method. We first derive the optimal solution of 68-landmark on 3D reconstruction, paving the way for the future derivation of other landmark methods on various image processing.
|
