Hello
There seems to be some problem with the ground truth. We have found several images where the angles that were provided seem quite incorrect.
For example:
1) Sample_16_295 pitch = 4.33400, yaw = 7.16285, roll = 3.60563
the person is looking right and up and the angles show that the person is looking almost ahead
2) Sample_13_058 pitch = 85.28499, yaw = -42.39845, roll = -44.37128
Here the pitch angle is quite big, but person is looking left (neither up nor down)
3) Sample_22_490 pitch= 63.6119 yaw= -19.0517 roll= -55.7710
and
4) Sample_05_592 pitch= -41.6929 yaw= 21.9899 roll= 4.3818
In both cases the person is looking down but in (3) the pitch angle is POSITIVE while in (4) the pitch angle is NEGATIVE. Roll angle in (3) also seems strange.
5) Sample_04_209 pitch= -60.7876 yaw= 44.1300 roll= 14.8411
and
6) Sample_01_394 pitch= 78.1949 yaw= -31.8000 roll= -77.2456
In both cases the person is looking up but the pitch angle in (5) is NEGATIVE while in (6) the pitch angle is POSITIVE. Roll angle in (6) is also strange.
7) Sample_22_106 pitch= -76.2816 yaw= -9.0325 roll= -66.3407
the person is looking almost ahead but pitch and roll angles are big
These are just some examples to illustrate the problem. We've identified about 100 already (while we've checked only a fraction of the data).
Would you mind having a look at the examples above and double-checking that there is no mistake in the provided angles? Or may be we misunderstood something?
Thanks in advance
Posted by: Dmytro @ Jan. 26, 2017, 3:25 p.m.Hi
Thanks for your message. We are investigating on it.
Best
Shahab
Hi!
I want to thank you for pointing out this mistake. As I got the angles from the rotation matrix they are bot uniquely defined. However I did not check all of the labels when generating this data for the competition. I am currently going over it and will let you know once it is all in order. I am truly very sorry.
Posted by: icv @ Jan. 31, 2017, 7:12 a.m.regarding to our last question, about mismatch angles. Did you check it? Do you have any information?
Posted by: Dmytro @ Feb. 8, 2017, 11:21 a.m.HI!
Yes, I have checked them and I have added a file called new_angles.csv to the drive. If you have eny more questions, please ask.
Posted by: icv @ Feb. 8, 2017, 12:27 p.m.labels the same?
and what do empty lines mean?
Hi, empty lines are the missing angles as discussed in some other threads. Everything else is the same.
Posted by: icv @ Feb. 8, 2017, 2:12 p.m.Thank you
Posted by: Dmytro @ Feb. 8, 2017, 2:12 p.m.