search

Hydrogenion / ViPC

PyTorch implementation for View-Guided Point Cloud Completion
MIT License
47 stars 7 forks source link

How do you pretrain the modality transfer network? #2

Closed czvvd closed 2 years ago

czvvd commented 2 years ago

the loss function is CDloss of complete point cloud and reconstrusted point cloud?

Hydrogenion commented 2 years ago

CD loss + EMD loss are used to train reconstrusted and complete point cloud.

czvvd commented 2 years ago

Thanks a lot,but I wonder that what gpu do you use to train the model and how long do you need to train 100 epochs?Because the dataset is so big,it contains over 730k training sets.

Hydrogenion commented 2 years ago

Yeah, the dataset is so big, maybe you sample a small set for debugging your network. BTW, I use 2V100 (32GB) and 8GTX1080Ti(11GB) for trainging it. It takes about one week.

czvvd commented 2 years ago

Thanks a lot for releasing code, but it seems that the utlis folder still needs to be updated.

Hydrogenion commented 2 years ago

the utlis has been already committed!

duzhenjiang113 commented 2 years ago

Hi, could you please tell me how to pre-train the modality transfer network? Are you using the entire network for training? Do you only update the network parameters of modality transfer during training?