princeton-vl / Coupled-Iterative-Refinement

MIT License
104 stars 20 forks source link

Poses in the world coordinates #3

Closed shanice-l closed 2 years ago

shanice-l commented 2 years ago

Hi authors!

Thanks for your great work! I've noticed that you used the object poses in the world coordinates. Could you please tell me your consideration? Can I replace it with object poses in camera coordinates?

RSVP.

lahavlipson commented 2 years ago

Yes, you can use object poses in camera coordinates instead of world coordinates. Please note that you will have to swap variables ii and jj in the input to projective_transform, MoBA and BA accordingly. If you're using the Modified BD-PnP approach, you'll also need to invert the pose input to renderer.

We use world coordinates since this is the format expected by the BOP benchmark.

shanice-l commented 2 years ago

Oops, I understood it. You wrote P = K^(-1)KP in your code, and it's a little bit confusing.