search

huangjh-pub / di-fusion

[CVPR'21] [Jittor & Pytorch] DI-Fusion: Online Implicit 3D Reconstruction with Deep Priors
122 stars 13 forks source link

About the scannet training scripts #5

Closed chensjtu closed 2 years ago

chensjtu commented 2 years ago

Hey! Thanks for your wonderful and impressive work. I'm a newbie in computer graphics and I've tried to use the Di-Fusion to recon the RGBDs in scannet, but I encounter some problems in the data preprocessing. How can you pre-process the images as well as depths in scannet dataset? Many thanks for sharing the data! would you please tell me how to use di-fusion on scannet correctly?

watarungurunnn commented 2 years ago

@chensjtu Hey! Could you solve this problem? If so, could you send me the script? Thank you.

heiwang1997 commented 2 years ago

Hi guys, sorry for the late response. For ScanNet we first convert the sens files into image and depth sequences using the script from here. Then the depth maps are back-projected to point clouds using the intrinsic matrix. The RGB images are used only as post-processing to add textures to the reconstructed meshes.

Could you please share with me more specific problems you've met? I think that will help me better locate where you are struggling with :)

huangjh-pub commented 2 years ago

Closed due to inactivity. Feel free to reopen if needed.