SMPL Initialization

Question

SMPL Initialization

huanngzh opened this issue a year ago · 5 comments

Why is the depth map used to initialize a NeRF of canonical SMPL? Use color map rendered from mesh will be efficient.

Answer 1 · 2024-01-09T17:59:17.000Z

Hi, using color map to initialize NeRF brings unnecessary color priors, e.g. white foreground. We tend to initialize NeRF's geometry without introducing appearance preferences.

Answer 2 · 2024-01-09T18:56:00.000Z

Thank you so much for the threestudio implementation of DreamWaltz! We really appreciate it.

Answer 3 · 2024-01-10T05:36:56.000Z

Well, got it. Btw, I'm honored to implement such a good avatar generation work as an extension of threestudio.

Answer 4 · 2024-01-10T05:37:57.000Z

During the implementation process, it was discovered that when your code exported the mesh, depth, and pose maps of SMPL, the rendered mesh were inconsistent with the other two forms in the same camera input, as shown in the figure below.

Could you please let me submit a pull request to resolve this issue?

Answer 5 · 2024-01-10T06:44:30.000Z

Sure! Thanks for the bug report and please feel free to submit a pull request.