search

fxia22 / pointnet.pytorch

pytorch implementation for "PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation" https://arxiv.org/abs/1612.00593
MIT License
2.13k stars 629 forks source link

Different number of points #88

Open Amir-Abi opened 3 years ago

Amir-Abi commented 3 years ago

Hi, point net was originally trained on a fixed-sized input , and I wonder if there is a efficient way to deal with different numbers of points in a batch. for example, (3, 10) and (3, 11) are my samples, how I should use them in the same batch ??? is there any better way than padding & passing mask that it'd used in NLP ???

DJNing commented 2 years ago

I got same question here. But it seems pytorch requires same data shape when stacking into batches

jasmineytt commented 2 years ago

I got same question, how you solve the problem?

Med-Amine-saighi commented 1 year ago

I think you should implement a collate function and pass it to the DataLoader so you can get heterogonous batches