Currently fitting results are not very good even after manual scale and center alignment. More feature points and alignment automation are needed.
Use, for example, dlib for preprocessing, and push found landmarks as an image metadata for fitters.
On each step Fitter or Viewport should calculate the best transformation matrix.
Viewport may be used as a workaround, because Face model is calculated there at the moment, and it's actually a work for a proxy #64
Currently fitting results are not very good even after manual scale and center alignment. More feature points and alignment automation are needed.
Use, for example, dlib for preprocessing, and push found landmarks as an image metadata for fitters. On each step
Fitter
orViewport
should calculate the best transformation matrix.Viewport
may be used as a workaround, becauseFace
model is calculated there at the moment, and it's actually a work for a proxy #64