-
Notifications
You must be signed in to change notification settings - Fork 38
Description
Thank you for your released code. Recently, I'm doing a project using the Objectron dataset. Now, I need obtain the GT rotation matrix and euler angles labels of each object. The given labels of Objectron contain the possible answer of GT rotation matrix (either camera.transform or camera.view_matrix). This is also explained by an official related issue in Objectron.
Then, I utilized the camera.view_matrix multipling object.rotation for each instance as the GT rotation matrix, and wanted to further visualize the orientation of each object. I have to convert the GT rotation matrix into corresponding euler angles. Below is my code.
from scipy.spatial.transform import Rotation
rot_mat_2 = np.transpose(rot_mat) # rot_mat is the GT rotation matrix
euler_angles = Rotation.from_matrix(rot_mat_2).as_euler("zxy", degrees=True) # is this order right?
[roll, pitch, yaw] = euler_angles I plotted three euler angles for further checking. However, the orientation is not always visually right. Some examples are shown below.
positive / right examples:
![]() |
![]() |
![]() |
negative / wrong examples:
![]() |
![]() |
![]() |
I'm not sure if this is my fault or just the GT labels of orientation have large noise caused by human annotators. I noticed that you have shown right visualiztion of object orientation by euler angles in your paper. But I cannot find the corresponding code in this repo. Could you please share your processing steps? Thank you very much.





