search

wutong16 / Voxurf

[ ICLR 2023 Spotlight ] Pytorch implementation for "Voxurf: Voxel-based Efficient and Accurate Neural Surface Reconstruction"
Other
399 stars 28 forks source link

use dvgo init for the w/ mask setting #4

Closed zhirui-gao closed 1 year ago

zhirui-gao commented 1 year ago

Thanks for your awesome work. For the w/ mask setting, i noticed that you use density voxel grid instead of sdf voxel grid. Could you explain why?

wutong16 commented 1 year ago

Hi. For the two-stage Voxurf training, we use the SDF voxel grid for both w/ and w/o mask settings. In w/ mask, specifically, we use an additional step of DVGO initialization (as mentioned in the paper, Sec. 5.1), which helps construct a mask cache for better sampling. The density voxel grid is used in this step for convenience, only for building the mask cache.

zhirui-gao commented 1 year ago

I see. Thank you!