Voxel Files Outnumber Point Cloud and Label Files in SemanticKITTI Sequences — Causing Mismatched Data Entries

[xyxiongcode]

I'm currently working with the SemanticKITTI dataset for 3D occupancy prediction training , and I’ve encountered a data mismatch issue regarding the voxels subfolder.

Problem Description
In multiple sequences (e.g., sequences/00, sequences/01, etc.), the number of .npz voxel files is significantly greater than the number of corresponding .bin (velodyne) and .label files. Specifically:

Some voxel files (e.g., 002480.npz) exist in the voxels directory, but the corresponding 002480.bin and 002480.label do not exist in velodyne/ or labels/.

As a result, when training or preprocessing data that depends on synchronized inputs (voxels + lidar + labels), this inconsistency can lead to errors or ignored samples.

This also creates difficulty in aligning data for supervised learning or evaluation steps where perfect correspondence between modalities is required.

[jbehley]

Please verify that you downloaded (and also extracted) the complete data. It might be that you simply hadn't have enough disk space and some of the .bin and .label files have not been extracted?

[xyxiongcode]

Thank u very much fore ur detailed reply, i have solved this problem