search

vitoralbiero / img2pose

The official PyTorch implementation of img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation - CVPR 2021
Other
588 stars 109 forks source link

Questions about pose pred #56

Closed eugeneYz closed 2 years ago

eugeneYz commented 2 years ago

1.What is the output measurement unit :pose_pred: [array([ 0.14125733, -0.02719707, -0.01702601, 0.0928278 , 0.5363695 ,12.56241608])] rx, ry,rz in rad and tx ty tz in cm?

  1. What is the center of pose ? Is the pix center of box? Thank you!
eugeneYz commented 2 years ago

‘Cause I tried to test the pose_pred on my web cam and found the result of translation in z-axis a huge error.Does it relate to the camera intrinsic? Looking forward to your comment!

vitoralbiero commented 2 years ago

The rotation output is in radians, and the translation is an arbitrary value. If you want to read more about the output, it is the same format as OpenCV's SolvePnP.

The camera intrinsic should be defined in the same way as we do, if you changed that, you will not get an accurate prediction.

Hope this helps.

eugeneYz commented 2 years ago

Sorry,I still confused about the head translation an arbitrary value.What if the output be [array([ 0.14125733, -0.02719707, -0.01702601, 0.0928278 , 0.5363695 ,12.56241608]) . What is the unit of lenth?