Closed Eby-123 closed 4 years ago
Hi, we currently dress SMPL model with the garments. To get better faces you will have to register SMPL model with faces and use that as the underlying body.
Thanks a lot for your reply.Besides, I still have a little confusion.Would you explain the confusion for me?
1 Is the pose of the 3D human model output by MGN the same as that of the human body in the input image?
The figure above shows the human posture of MGN's input image.(Figure 1)
The figure above shows the postures predicted by MGN,(Figure 2)
As you mentioned, figure 1 and Figure 2 should have the same body posture. But now figure 1 and Figure 2 have different body postures. Why?
MGN predicts garments and body (in T-pose) and SMPL pose parameters. You can use the code 'dress_SMPL.py' to repose the garments according to pose predicted by MGN.
Thank you for your reply. It helps me a lot.
In addition, how to generate Obj files ('pants. Obj ',' shortpants. Obj ',' shirtnocoat. Obj ',' tshirtnocoat. Obj ',' longcoat. Obj ') for five types of clothes?
How to generate obj file of shoes or hair?
Any suggestion will help me a lot, thank you.
@bharat-b7 Hi, I've run the file 'dress_SMPL.py' successfully,and the result amazes me very much. But, the body in the results lacks the realistic face.How to re-target garments including both texture and geometry which can make the results more realistic? Any advice would be appreciated.
Hello, bother you, I want to know how you installed the mesh when you installed it. The mesh link URL does not support py27. Then when I found the mesh tested on py27, the installation also had problems. This problem bothered me I really need your help, thank you very much
copying mesh/topology/decimation.py -> build/lib.linux-x86_64-3.5/psbody/mesh/topology
copying mesh/topology/connectivity.py -> build/lib.linux-x86_64-3.5/psbody/mesh/topology
copying mesh/topology/init.py -> build/lib.linux-x86_64-3.5/psbody/mesh/topology
copying mesh/topology/linear_mesh_transform.py -> build/lib.linux-x86_64-3.5/psbody/mesh/topology
copying mesh/topology/subdivision.py -> build/lib.linux-x86_64-3.5/psbody/mesh/topology
creating build/lib.linux-x86_64-3.5/psbody/mesh/geometry
copying mesh/geometry/rodrigues.py -> build/lib.linux-x86_64-3.5/psbody/mesh/geometry
copying mesh/geometry/triangle_area.py -> build/lib.linux-x86_64-3.5/psbody/mesh/geometry
copying mesh/geometry/vert_normals.py -> build/lib.linux-x86_64-3.5/psbody/mesh/geometry
copying mesh/geometry/tri_normals.py -> build/lib.linux-x86_64-3.5/psbody/mesh/geometry
copying mesh/geometry/init.py -> build/lib.linux-x86_64-3.5/psbody/mesh/geometry
copying mesh/geometry/cross_product.py -> build/lib.linux-x86_64-3.5/psbody/mesh/geometry
copying mesh/geometry/barycentric_coordinates_of_projection.py -> build/lib.linux-x86_64-3.5/psbody/mesh/geometry
creating build/lib.linux-x86_64-3.5/psbody/mesh/serialization
copying mesh/serialization/serialization.py -> build/lib.linux-x86_64-3.5/psbody/mesh/serialization
copying mesh/serialization/init.py -> build/lib.linux-x86_64-3.5/psbody/mesh/serialization
running build_ext
[CGAL] deflating cgal from "mesh/thirdparty/CGAL-4.7.tar.gz" to "/home/ybb/study/MGN/mesh-master/build/temp.linux-x86_64-3.5"
building 'psbody.mesh.aabb_normals' extension
Traceback (most recent call last):
File "setup.py", line 266, in
It seems mesh library is no longer supported for python 2.7 Raise an issue at https://github.com/MPI-IS/mesh
The code to dress SMPL works with python 3 as well. We have tested MGN with python 2.7 but if you update dependencies such as dirt, cPickle etc., this repo should work with python 3 as well.
The code to dress SMPL works with python 3 as well. We have tested MGN with python 2.7 but if you update dependencies such as dirt, cPickle etc., this repo should work with python 3 as well.
Confirmed. I got it working with python3.7.
@bharat-b7 Hi. Is there any plan to release the train code? Thanks!
MGN predicts garments and body (in T-pose) and SMPL pose parameters. You can use the code 'dress_SMPL.py' to repose the garments according to pose predicted by MGN.
@bharat-b7 Seems like the repose is just using the displacement map rather than using the pose parameter.
@bharat-b7 Hi. Is there any plan to release the train code? Thanks!
@Frank-Dz , As mentioned in another issue comment, look (grep
) for def train
in network/base_network.py
. On my machine, it's line 431. Unfortunately I don't know details of how to use this yet. You can read BLB's paper. Hope that helps! :smile:
Hi, we currently dress SMPL model with the garments. To get better faces you will have to register SMPL model with faces and use that as the underlying body.
@bharat-b7 @Eby-123 can you please let me know how to do that.Will it also create hair from input images
@bharat-b7 Hi, I've run the file 'dress_SMPL.py' successfully,and the result amazes me very much. But, the body in the results lacks the realistic face.How to re-target garments including both texture and geometry which can make the results more realistic? Any advice would be appreciated.