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

issue of debug #71

Closed MY-CODE-1981 closed 3 years ago

MY-CODE-1981 commented 3 years ago

The explanation on pointnet2+ is not clear.

ethnhe commented 3 years ago

Please just talk about technique and code on GitHub, not personal attacking :-). This project is open-sourced for free under my hobbies and hoping to help people. That also means I can only try my best to maintain it during my spare time. So everyone is welcome to help, contribute and further improve the project in the GitHub community.

About the fixes, Step1) and Step2) are needed if you built the extension in Pointnet2_PyTorch. If you use PVN3D/setup.py to built the extension following README, a copy of the built _ext*.so will be in /PVN3D/pvn3d/lib/pointnet2_utils/init.py and the program will work well.

Step 3) is a good fix to train_ycb_pvn3d.py in Pytorch-1.5 branch and I will update it.

About Step 4), you can adjust the batch size according to your GPU capability. But cutting the batch size may affect the performance of the trained model as we use Batchnorm in our model.

Anyway, thanks for the contributions.

ethnhe commented 3 years ago

Information of our GPU setting can be seen in #10