search

ethnhe / PVN3D

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

diameter of YCB models #88

Closed shockjiang closed 3 years ago

shockjiang commented 3 years ago

Hi, I'm reading the code and try to re-produce your results. As your paper said, AUC is calculated based on the offset distance less than 10 percents of its diameter. However, when the code calculate the AUC, it seems that a fixed value is used (0.1):

Does I got it right, or I missed something?

Thank you.

shockjiang commented 3 years ago

@ethnhe could you leave me a quick response. I'm kinda hurry to this work.

shockjiang commented 3 years ago

@ethnhe any suggestion?

ethnhe commented 3 years ago

As illustrated in our paper, the AUC on the YCB-Video is obtained by varying the distance threshold following PoseCNN[52] and DenseFusion[50], the maximum threshold of which is 0.1. Our python code is consistent with their Matlab code.

For the metric on the LineMOD dataset, we report the accuracy of distance less than 10% of the object's diameter (ADD-0.1d) following Model based training, detection and pose estimation of texture-less 3d objects in heavily cluttered scenes. the same setting in all other baselines in the table.