NVlabs / BundleSDF

[CVPR 2023] BundleSDF: Neural 6-DoF Tracking and 3D Reconstruction of Unknown Objects
https://bundlesdf.github.io/
Other
992 stars 109 forks source link

Is ground-truth required during training? #162

Closed lpy523 closed 3 weeks ago

lpy523 commented 1 month ago

I noticed that the loss function is mentioned in the paper in both the Online Pose Graph Optimization and Neural Object Field sections, and my question is whether ground-truth pose data is needed to supervise the model throughout its training?

I'm sorry I don't understand the paper well enough, if someone could answer my question, I would greatly appreciate it!

wenbowen123 commented 3 weeks ago

No it's not needed. It's only a loss for optimization, w/o comparing to GT.

lpy523 commented 2 weeks ago

Where specifically in the code is the loss function in the Online Pose Graph Optimization section defined and called? Thanks! @wenbowen123