search

patrikhuber / superviseddescent

C++11 implementation of the supervised descent optimisation method
http://patrikhuber.github.io/superviseddescent/
Apache License 2.0
402 stars 188 forks source link

pose estimate #10

Closed songminglong closed 9 years ago

songminglong commented 9 years ago

@patrikhuber , hi , I want to know that how do you get 3-d face coordinates of "Mat facemodel;"(in pose_estimation.cpp), and three angles(pitch,yaw,roll) of "Mat predicted_params" are relative to given 3-d face, is it right?

patrikhuber commented 9 years ago

@shangguanxiaohu Hi, The face coordinates in the example are just from our internal 3D model, but you can use any face mesh basically. We are actually planning to release our 3D model stuff too, in the next few weeks. The angles are in a coordinate system however you choose to learn it at training-time. So yes, you can see it as relative to the 3D face.

songminglong commented 9 years ago

I think given 3-d face should be more frontal, and there are so many parameters of the model(eg. focalLength, tz and other params) that may reduce the stability of the model. Actually, I have trained the pose estimation model use your given faceModel data, and found that estimated angles have large changes in test(I have normalize landmarks to the center(500,500)).

patrikhuber commented 9 years ago

What do you mean with "given 3-d face should be more frontal"? It should be frontal - is it not? If you fix the focal length (i.e. assuming an affine camera model, if you set it large), there's 6 parameters. I don't think that's very much? I would say, try with a fixed focal length, and then go from there!

patrikhuber commented 9 years ago

I'm going to close this, we can always reopen the issue if you have more questions or problems.