Tangshitao / MVDiffusion_plusplus

MVDiffusion++: A Dense High-resolution Multi-view Diffusion Model for Single or Sparse-view 3D Object Reconstruction
138 stars 3 forks source link

Question about the paper #2

Closed Gynjn closed 6 months ago

Gynjn commented 6 months ago

Firstly, thanks for sharing your great work.

In page 7, I guess the generation mask should be 0? not 1?

Also I understand that the conditional mask is value 1 for all, and the generation mask is value 0 for all. Then how did you use the mask provided by the datasets?

Tangshitao commented 6 months ago

We use the object mask in the vae training and also the neus training. The mask is different from the mask we use in the diffusion model. The mask used in diffusion model is mainly for differentiate the conditional branch and generation branch and it's either all 0 or 1.