Image with hand, but no pose_m

[hajiejue]

thank you for your work!
I met a question about datasets.
For some images, the hand is visible but pose_m is all zeros, did I miss something?

[ychao-nvidia]

That is expected. Some of the images do not have hand pose annotation. pose_m will be all zeros for those cases. Those images have been excluded from evaluation.

There is a note for that in this section.

[hajiejue]

Thank you for your quick reply.
Why we have hands but without label?

[ychao-nvidia]

That is an artifact from how we generated the ground-truth annotation. If the number of labeled keypoints from the all views is below some threshold, we skip the annotation for that frame since the ground truths may not be accurate.