search

TRI-ML / packnet-sfm

TRI-ML Monocular Depth Estimation Repository
https://tri-ml.github.io/packnet-sfm/
MIT License
1.23k stars 242 forks source link

question about camera reconstruct and project function #187

Closed livey closed 2 years ago

livey commented 2 years ago

At camera.reconstruct, it uses self.Twc @ Xc, and at camera.project, it uses self.Tcw @ X. However, Twc means from world to camera translation, and Tcw means from camera to world translation, should it be self.Tcw @Xc and self.Twc @ X respectively?

VitorGuizilini-TRI commented 2 years ago

Thank you for that, I'll double check. Since everything works fine, worst case scenario is that the names are swapped, so it's just a matter of renaming them.

livey commented 2 years ago

Yes. It does not influence the results. But the pose prediction has a different meaning.

akshay-ka commented 1 year ago

Yes. It does not influence the results. But the pose prediction has a different meaning.

So, this should make it clear that the PoseNet output corresponds to a transformation from the "source image" to "target Image". (Target image = input image that is fed to the depth network)

@livey This is also what you mean with your comment "But the pose prediction has a different meaning." right ?