Difference in Dataset Format

[zhiminc]

Thanks for the great work! I find the the format of dataset download from hugging face is different from the format using in the training stage (npy in hugging face and npz in the dataloader). The shape of data in the npy file are N, N, and (B+1)x512x256x32, which is totally different from the npz format:

input: (B+1)x512x256x32 matrix # B is the number of previous frames, default is 10
label: (B+1)x512x256x32 matrix # B is the number of future frames, default is 10
invalid: (B+1)x512x256x32 matrix # same as label

May you tell me what's the meaning of the data in the npy files and the difference to the npz file? Thanks a lot!

[zhiminc]

I find that in the code of data preprocessing, only the npy format is stored in the voxel folder. I cannot find any code to generate npz voxel file that is utilized in the following dataloader code:

for i in self.split:
sample_path = glob(os.path.join(self.root_dir, 'voxel', '{:04d}'.format(i), '*.npz'))
self.data_path.extend(sample_path)