Open franciszzj opened 5 years ago
@Franciszzj As far as I can tell, you cannot actually do this. There are two reasons: (1) The uv-map from the Basel face model collapses many vertices on top of each other when you flatten out a mesh from 300W-LP to get a target position map, which makes it impossible to predict different locations for these vertices, and (2) The 53215 vertex count includes predictions for the neck region, which PRNet does not make. The paper doesn't really describe the different submeshes of the Basel face model that they are using, but the main point is that the uv-map gives floating point coordinates, but we are setting up a correspondence where vertices are pixels with integer values in [256,256]
@Franciszzj As far as I can tell, you cannot actually do this. There are two reasons: (1) The uv-map from the Basel face model collapses many vertices on top of each other when you flatten out a mesh from 300W-LP to get a target position map, which makes it impossible to predict different locations for these vertices, and (2) The 53215 vertex count includes predictions for the neck region, which PRNet does not make. The paper doesn't really describe the different submeshes of the Basel face model that they are using, but the main point is that the uv-map gives floating point coordinates, but we are setting up a correspondence where vertices are pixels with integer values in [256,256]
Hi @wungemach, do you know any easy method to remove neck regions when I change the resolution of UV position map...
The 3DMM output mesh has 53215 vertices. How to use PRNet to generate 53215 vertices mesh? Where can I find/generate the corresponding index? The index given (face_inds.txt) can generate only 43867 vertices.