search

SamsungLabs / NeuralHaircut

Neural Haircut: Prior-Guided Strand-Based Hair Reconstruction. ICCV 2023
Other
532 stars 48 forks source link

FLAME fitting #30

Open Suho0922 opened 3 weeks ago

Suho0922 commented 3 weeks ago

I have some questions about FLAME fitting.

I performed FLAME mesh fitting on the H3DS dataset and my own dataset by following the instructions for preparing a custom dataset. However, some of the FLAME fitting results were not satisfactory (the FLAME meshes do not match the original scenes well).

  1. In the Pixie results, the fitting for back poses was not accurate because the face could not be detected. Should I include these back pose Pixie results for the initialization of FLAME fitting?

  2. How accurate does the FLAME fitting need to be for the H3DS dataset? I’m unsure what level of fitting accuracy is considered acceptable for the next steps.

  3. Should I need DECA model for pixie?

  4. Do you have any advice on how to improve FLAME fitting quality? For example, is there a recommended way to select good views for initialization, or are there specific settings or strategies that can enhance the fitting accuracy?

Thank you for reading. :)

kkxt0802 commented 2 weeks ago

same question ，how to get the correct head_prior.obj