search

ethnhe / PVN3D

Code for "PVN3D: A Deep Point-wise 3D Keypoints Hough Voting Network for 6DoF Pose Estimation", CVPR 2020
MIT License
482 stars 105 forks source link

A question about code, please help! #73

Closed 74284853 closed 3 years ago

74284853 commented 3 years ago

I recently tried to run your code, encountered a problem, I hope I can get your help, can your code output the six dimensional coordinates of the key point or the center point of the forecast target??

74284853 commented 3 years ago

I used the linemod dataset.

ethnhe commented 3 years ago

What're the six-dimensional coordinates mean? You can print the 3-dimensional coordinate in the camera coordinate system in the data preprocessing scripts: center point, keypoints.

74284853 commented 3 years ago

What're the six-dimensional coordinates mean? You can print the 3-dimensional coordinate in the camera coordinate system in the data preprocessing scripts: center point, keypoints.

Shouldn't the result predicted by the code include the translation coordinate T and rotation coordinate R of the object?

74284853 commented 3 years ago

For example, can I use your code to detect the actual coordinates of objects on the desktop, including translation coordinates (x, y, z) and rotation coordinates (Rx, Ry, Rz)?

ethnhe commented 3 years ago

We output the transform matrix that can transform the object from its object coordinate system to the camera coordinate system, they are at lib/utils/pvn3d_eval_utils.py(link), where pred_RT[:3, :3] is the 3x3 rotation matrix and pred_RT[:3, 3] the 3x1 translation vector.