facemesh canonical face model and uv visualization

[hiukim]

Looks like the face mesh model in the new mediapipe solution has been updated. e.g. There were 468 landmarks before, and now there is 478.

I would be interested to get the canonical face model as well as a uv visualization map. The old data can be found here: https://github.com/google/mediapipe/tree/master/mediapipe/modules/face_geometry/data

But I couldn't find corresponding materials for the new version. Can you provide those? Thanks a lot.

[lucademenego99]

Any updates on this? Thanks

[emrdgn98]

Is there anybody who knows that? If you have any information about that reply pls. Also is there any way to get 3D points from the new model? Thanks.

[emrdgn98]

Hi guys, I found something which I want to share. You know Mediapipe released a new face mesh model which provides a face transformation matrix. This matrix includes a rotation matrix and a translation vector. That means if we know intrinsic parameters(it can be obtained with https://docs.opencv.org/4.x/dc/dbb/tutorial_py_calibration.html) and transformation matrix we can find 3D points for each point on the face mesh with the reprojection method. Am I right? If you have any information about that please reply

[hiukim]

I figured the landmarks are the same as the old model. The 10 extra landmarks are for the iris (newly added)

Not sure they are using the same canonical face model or not. Hope that someone can confirm.