Human IDs are different in different images?

[cherryjm]

I get the human id as

GTA-IM-Dataset/vis_skeleton_pcd.py

Lines 62 to 64 in 31d7baa

	human_id = id_map[
	np.clip(int(p[1]), 0, 1079), np.clip(int(p[0]), 0, 1919)
	]

It is the id at the head location.

It seems that the human id is not unique. Or is it because the head is occluded or out of the field of view?

[andreaskoelsch]

I second your problem. This approach of taking the color at the head location is not very stable.

1. It does not work if the head is occluded, as for example in FPS-5/2020-06-09-16-09-56/00067*
2. It does not work if the position of the head is incorrect, cf. joints_2d are not exact #6

[ZheC]

Yes, the human IDs are different across sequences. To obtain the human ID, we have to use the head position in the image. While this works for most cases, it will have issues when the head is occluded as pointed by @koelscha. A more robust way would be checking the ID of all human joint positions, and use the major one.