Inquiries About Challenges

I have some questions regarding the parts presented in the bins for each phase:

  1. Are we trying to estimate the pose of arbitrary parts placed in a bin? (i.e. parts that may not have been in datasets.) It seems like the datasets just show example parts that could appear.

  2. Do we have to express the pose estimates with respect to the camera frame?

  3. Is there a specific number of parts we need to estimate the pose of in each robot cycle?

Thank you kindly,
Dave

  1. the list of parts is fixed - there are 10 objects in Phase 1 of the dataset. Each object may have multiple instances present in the scene. The list of objects present in the scene is available for each scene. I.e. you know, which parts you are looking for in each scene.
  2. Pose estimates should be wrt the world, which also matches with Photoneo pose.
  3. You need to detect all possible instances for the provided list of parts

As a follow up question, are the models files of the parts represented in their zero orientations?

Thank you kindly,
Dave

Yes, that is correct!