Is the training_data.txt the label file? I found that the number of the rows does not match the number of training samples, is there something wrong? Can you introduce the meaning of the digits in each line?
Posted by: bnu_hp @ Jan. 12, 2017, 12:17 p.m.Hi! I will look into it and reply as soon as possible.
Posted by: icv @ Jan. 12, 2017, 2:56 p.m.Hi, I have included a README file in the google drive folder with explanations.
Posted by: icv @ Jan. 12, 2017, 4:18 p.m.Hi Sir,
I have some questions regarding the training data groundtruth.
Firstly, I want to clarify if the angle values in training_data.txt of head pose ( the last 3 values ) is of yaw, pitch, roll order?
Also, there are some samples where their yaw, pitch, roll are estimated to be 5732.484 and nose location is (inf,inf,inf), thus, are those samples all estimated wrongly ?
Thank you!
Posted by: justin0605 @ Jan. 23, 2017, 10:14 a.m.Hi!
Those samples are indeed estimated wrongly. The kinect could not calculate the nose location, thus also giving an insane angle. Sorry for not clarifying. I actually thought I had removed such instances.
However as for the angles, I am terribly sorry. It should be pitch, yaw, roll. I will also fix it on the competition page.
Posted by: icv @ Jan. 25, 2017, 11:33 a.m.