search

modelscope / richdreamer

Live Demo：https://modelscope.cn/studios/Damo_XR_Lab/3D_AIGC
https://aigc3d.github.io/richdreamer/
Apache License 2.0
362 stars 13 forks source link

Question about Input Data Choices for Stable Diffusion in DMTet and NeRF Methods #40

Closed DaLi-Jack closed 4 days ago

DaLi-Jack commented 4 days ago

Hi! Great work! However, I have a question. Why is the normal map used as input for the stable diffusion model in the DMTet-based method, while the RGB map is used in the NeRF-based method? Is there a specific reason for this? Looking forward to your response!

lingtengqiu commented 4 days ago

For NeRF, since the initial normal map produced by NeRF is not as good as the one generated by DMTet, we use a different approach. Specifically, we use an RGB map as the initial input for Stable Diffusion. After achieving stable training, we switch to using the normal.

DaLi-Jack commented 4 days ago

Thank you for your reply, it has resolved my problem!