You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The Pulsar renderer is currently the fastest way to render pointclouds using pytorch3d. However, it only supports one focal length.
Motivation
It would be nice to support two focal lengths so that we can do inference with calibrated cameras from real life, which often have different fx and fy. This would enable some pose refinement / SLAM applications.
The text was updated successfully, but these errors were encountered:
We aren't planning to work on this kind of thing soon. Even fixing and merging #1369 is not currently being worked on. Maybe you or someone else might do this?
Thanks. Can't work on that fix you mentioned currently, but for anyone else seeing this, here is a workaround.
Use the Pulsar renderer as an idealized virtual camera plane with some fixed focal length and size. Then the transform from the virtual camera plane to a focal-length corrected camera image is simply a homogeneous transform, which can be achieved by F.grid_sample.
🚀 Feature
The Pulsar renderer is currently the fastest way to render pointclouds using pytorch3d. However, it only supports one focal length.
Motivation
It would be nice to support two focal lengths so that we can do inference with calibrated cameras from real life, which often have different fx and fy. This would enable some pose refinement / SLAM applications.
The text was updated successfully, but these errors were encountered: