youmi-zym/GO-SLAM

Was the tracking process using the mapping result?

nnop opened this issue · 4 comments

nnop commented

Great work and amazing results!
I have a question about how the mapping result, which is the implicit surface, was used for tracking.
I didn't find that in the paper.

Hi,

We didn't let the results from mapping feedback to tracking. The Mapping is only used for better visualization, i.e., mesh rather than point cloud. But you can take a try to build the interaction between mapping and tracking.

Best,
Youmin

nnop commented

In that way, the method is just a simple loosely combining of DROID-SLAM and neural surface reconstruction?

Hi,

No, the main problem we want to solve is how to build globally consistent 3D reconstruction. So, the key is how to globally optimize the pose graph and instantly update the 3D map.

Best,
Youmin

nnop commented

Thanks for your explaination.