search

ethnhe / PVN3D

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

Segmentation Results #60

Closed hillaric closed 3 years ago

hillaric commented 3 years ago

Hello, ethnhe. I am very interested in your work, it is very effective. In fact, I have some questions in PVN3D segmentation results. In table 7, you compare your segmentation result with Mask-RCNN, but the miou of 2D is the same as miou of 3D?

ethnhe commented 3 years ago

The results reported in the paper are all in '3D'. Specifically, for Mask-RCNN, we projected the sampled point cloud fed into PVN3D back to the 2D image and get the corresponding label predicted by Mask-RCNN to calculate the mIoU. Since the sampled point cloud is randomly selected from the whole image, the distribution of labels on the point cloud is close to the RGB image.

hillaric commented 3 years ago

thanks.