number of parametrized mesh faces unequal to SMPL mesh faces

[VVingerfly]

Hi,

First of all, thank you for making such a excellent work public available.

I downloaded the uv data and found the number of faces in the data is 13774. But there are 13776 faces in the SMPL mesh, which has two more faces.

I wonder if there is anything I missed? How to make a correspondence between the uv coordinates and SMPL mesh vertices?

Thanks a lot!

[GianKiMoon]

Hi, did you found out how to make the correspondence between the uv coordinates and SMPL mesh vertices? Thanks for your help!

[VVingerfly]

@GianKiMoon hi, the mesh DensePose provided has 2 missing triangles on the left and right foot thumb. The correspondence is provided by DensePose.

[zhhezhhe]

@VVingerfly Where did you find the information about the UV_Processed.mat? Do you know the meaning of "All_Faces" and "All_FaceIndices"? Thanks for your help!

[VVingerfly]

@zhhezhhe ‘All_Faces’ means faces of 2d uv mesh. 'All_FaceIndices' means each face index from 1 to 24 corresponding to 24 body parts.

[Rhyssiyan]

@VVingerfly I have some confusion. I found there are 7829 vertices in the 'All_vertices' key and there are only 6890 vertices in the SMPL model. So Are there overlaps between different parts? Thanks for your help.

[zhhezhhe]

@Rhyssiyan I think there are overlaps only in the same part. There are no overlaps between different parts. Am I right.

[Rhyssiyan]

Hi @zhhezhhe , thanks for the discussion. I think there exists overlaps between different parts. There is a case in the demo data. I will take vertice 3073 in the SMPL model as an example.

The vertices 6493,7810 in ALP_UV['All_vertices'] correspond to the same vertice 3073 for the demo data.

From the above figure, we can see that the faces are different for the vertice 6493 and 7810. So I think there are overlaps between different parts.

[zhhezhhe]

@Rhyssiyan I think you are right. It make sense.

[VVingerfly]

@Rhyssiyan Hi, 'All_vertices' are the vertices of the 24 parts of the original SMPL mesh, so there will be more vertices in the seams after you cut the original closed mesh.

[Rhyssiyan]

Thanks for your helpful reply. I have another question. Is 'All_vertices' obtained by cutting the mesh in 'basicmodel_m_lbs_10_207_0_v1.0.0.pkl'? In other words, Isn't pose information included in the ground truth uv map for densepose? What we regress is the coordinates for the standard T-pose. Is it true?

[VVingerfly]

@Rhyssiyan 'All_vertices' is obtained by cutting the T-pose mesh.
DensePose regress the uv coordinates of the pixels in the color image, which are correspondent to the 3d vertex of the T-pose mesh.

[Rhyssiyan]

Thank you! It solves my confusion.