3D face reconstruction

[codediner]

Hi Anpei,

    Thank you for sharing your great work. and I am interested in the process of reconstructing 3D face.

    Could you share with me how you produce these amazing 3d face models for training?
    According to my knowledgement, you have used light stage technique but the cameras you used are far less than that. 
    So how could you produce high quality 3d face model? Do the models have some holes or something like bubble as noises on face?
    Hope for your answers.

Alex

[apchenstu]

For the face reconstruction, please reference to our previous paper: Sparse Photometric 3D Face Reconstruction Guided by Morphable Models

[codediner]

Hi Anpei, Thank you for your answer, I have read your paper, you have compared the quality with the paper Multiview Photometric Stereo using Planar Mesh Parameterization. For the reconstruction software, have you used SFM+MVS as theirs or some business software? Alex

[LansburyCH]

In the paper "Sparse Photometric 3D Face Reconstruction Guided by Morphable Models", we do not have multi-view inputs and therefore only use photometric stereo (PS) for reconstruction. But for the dataset we create for this paper, we combine MVS with PS for even better accuracy.

For the SFM+MVS part, we use Agisoft Photoscan to reconstruct. And then we fuse the MVS result with the normal computed from photometric stereo using Poisson integration with depth prior. We did not use any techniques from the paper "Multiview Photometric Stereo using Planar Mesh Parameterization".