search

QianyiWu / objsdf

:t-rex: [ECCV‘22] Pytorch implementation of 'Object-Compositional Neural Implicit Surfaces'
183 stars 5 forks source link

About SDF bbox #10

Closed MontaEllis closed 1 year ago

MontaEllis commented 1 year ago

Hi Qianyi, I wonder why you use "SDF bbox" to clamp SDFs. Could you explain this? I haven't find any description in paper.

https://github.com/QianyiWu/objsdf/blob/cb4cbd1f98e0391ee2200f7018fc9efd1f5d16a5/code/model/network.py#L106

QianyiWu commented 1 year ago

Hi, The reason behind it is just to make sure that all rays should at least intersect with one surface in the scene. With the clamping operation, the SDF will eventually become zero on the sphere surface which could be treated as a panorama background during rendering.

This is a trick adopted from VolSDF. You can refer to the Sec B.3 in the supplementary of VolSDF for more details.

MontaEllis commented 1 year ago

Thanks a lot!