Why eikonal loss weight = 0?

deepseek-ai / DreamCraft3D

[ICLR 2024] Official implementation of DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior

https://mrtornado24.github.io/DreamCraft3D/

MIT License

1.88k stars 81 forks source link

Why eikonal loss weight = 0? #37

Closed wqdong8 closed 5 months ago

wqdong8 commented 6 months ago

hi, great work! I want to know why eikonal loss weight is set to 0, during the coarse-neus phase.

MrTornado24 commented 5 months ago

Hi, we find high eikonal loss weight can lead to distortions in the modeled surface. The model may produce surfaces that are too smooth or lack details. In our case, as the surface is initialized from NeRF, we empirically find we don't need such regularization.