yunhe20/Grad-PU

comparison with NeuralPoints

lisj575 opened this issue · 4 comments

lisj575 commented

Hi, thank you for your excellent work. I notice that you make a comparison with NeuralPoints on PU-GAN dataset, but they need normals as supervision which is unavailable in the dataset. So how did you test their method?

yunhe20 commented

You can estimate the normals, and then use the estimated results for supervision.

lisj575 commented

Thanks for your reply. Have you modified their code to test it on the PU-GAN dataset with their pre-trained model? I have been unable to reproduce their results, even when dividing the point cloud into patches. The quantitative results I obtained are close to that of PU-GAN.

kangya998 commented

You can estimate the normals, and then use the estimated results for supervision.

Hi,Congratulations on the groundbreaking progress in your work! Can you share the details of training Neps

yunhe20 commented

Please directly ask the authors of NeuralPoints for help if you meet any problem during the model training or testing.