The Unified NPY‑Format Virtual‑Point‑Cloud Extended KITTI Dataset for Multimodal 3D Object Detection

This is a virtual point cloud extended version of the KITTI dataset built based on the VirConv project (github.com/hailanyi/VirConv). In the original KITTI dataset, raw LiDAR point clouds are stored in bin format, while RGB virtual point clouds are saved as npy files. The inconsistency of data formats brings difficulties to multimodal joint processing.

In this dataset, original LiDAR point clouds are converted into npy format to unify storage format for real point clouds and virtual point clouds. It supports collaborative reading, preprocessing, and network feeding of two kinds of point clouds. The dataset contains 3712 official KITTI training and validation samples, as well as odometry samples with pseudo‑labels for semi‑supervision tasks.