Issue with generating pre-training dataset, ShapeNetFeat

SimingYan / MaskFeat3D

[ICLR 2024] "3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining"

5 stars 0 forks source link

Issue with generating pre-training dataset, ShapeNetFeat #2

Open mfarazi1991 opened 1 month ago

mfarazi1991 commented 1 month ago

Hi,

Thanks for releasing the code. I have a hard time knowing how I can generate the .npy files for the Shapenet dataset so that I can have the structure as mentioned for ShapeNetFeat. I downloaded different processed files from point-BERT, but those .npy files are different (already sampled). Can you upload a file regarding how to create this dataset? The pre-process.py is just for surface variation and it works fine. Thanks!

SimingYan commented 1 week ago

Hi,

Thanks for your interest! For the point cloud data, my suggestion is you could directly download the raw mesh data from official ShapeNet website and sample point cloud by yourself. I think our code just needs the point cloud data to be formatted as [50000, 6].