Why flip z axis of coord map?
Closed this issue · 3 comments
hankexin1314 commented
dataset.py line 335
coord_map = np.array(coord_map, dtype=np.float32) / 255
coord_map[:, :, 2] = 1 - coord_map[:, :, 2]
Saafke commented
@hankexin1314 @hughw19 I would also like to know the answer to this question.
hughw19 commented
Originally, there are some discrepancies between the coordinate frames in the GT and in the vision system. I had to manually correct the discrepancy using this line of code. If you correctly render the NOCS GT by yourself, you may be able to avoid those ugly code.
Best,
He
… On Aug 13, 2020, at 7:40 AM, Xavier Weber ***@***.*** ***@***.***>> wrote:
@hankexin1314 <https://github.com/hankexin1314> @hughw19 <https://github.com/hughw19> I would also like to know the answer to this question.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub <#16 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEXSGHZYVNHT3WFIGFR75WLSAP3MDANCNFSM4KCTW4RQ>.
Saafke commented
Ok, I rendered the NOCS maps exactly like in the paper. I also have symmetrical objects (around the y-axis (this is the axis pointing up, as you use XZY coordinate system)), but for some reason, my NOCS predictions give the wrong colors (e.g. the object is completely red. I'll try removing that line! Are there any discrepancies I should be aware of?