FaceMesh-with-Attention model conversion

[saeidn95]

Hello All,

I am following on the issue (#143)
I am looking at the outputs for the generated onnx model for the FaceMesh, and try to match it with the description at the MediaPipe Face landmark detection guide in (https://developers.google.com/mediapipe/solutions/vision/face_landmarker). The MediaPipe suggest an additional output (output_facial_transformation_matrixes). Would anyone know how to get this output. Further, the MediaPipe suggests existence of a static 3D Canonical Face Model (https://github.com/google/mediapipe/blob/master/docs/solutions/face_mesh.md). Does anyone know how to get this?