find out even in the image, there is no hand, but true labels show there is hand pose

[hajiejue]

thank you for your work! I met a question about dataset. when I look at the image(location: data/20200709-subject-01/20200709_141841/840412060917/color_000013.jpg ), I can not see any hand. But when I load the label = numpy.load('data/20200709-subject-01/20200709_141841/840412060917/labels_000013.npz') and the print the pose_m, I find out the label is not all 0 for hand. Did I miss something or there is other explain for that.

[ychao-nvidia]

For some images, we might have a valid hand annotation even if the hand is outside the image view. This is possible because the hand pose is solved from multiple views, and we can still obtain a pose annotation as long as the hand is observable in a subset of views.

For a quick test, you can check joint_2d in the label file. The 2D joint position should be outside the image boundary.