KITTI-360 street chunks as point clouds and Gaussian splats (GaussianWorld subset, 16 chunks)

pl.load("hf_kitti360_gs_subset") loads this dataset. Sixteen chunks of KITTI-360 sequence 0000 pre-processed by GaussianWorld: per chunk the accumulated LiDAR point cloud (pc_coord.npy, ~3 M points, with segment ids) and a fitted Gaussian splat stored as separate arrays (coord, color, opacity, scale, quat). Coordinates are in the KITTI-360 world frame with offsets above 1 000 m - keep float64. This is the dataset the pl.load feasibility test of 2026-10-03 ran on (one chunk of 43 MB fetched in 5.5 s, ground removal unchanged).

import pardesline as pl

ds = pl.load("hf_kitti360_gs_subset")
print(ds.describe())  # roles, frames, units, labels, licence, citation
s = ds[0]
obj = s.points  # fetched now, cached under ~/.cache/pardesline/datasets/hf_kitti360_gs_subset

Provenance and licence

Source	Hugging Face Hub repo GaussianWorld/kitti_360_subset_chunk_gs_preprocessed, paths test
Scheme	hf
Size of the whole source	2.4 GB
Licence	CC-BY-NC-SA-3.0
Commercial use	not allowed (non-commercial licence)
Redistribution	forbidden
Licence note	The Hub card declares no licence; KITTI-360 upstream is CC BY-NC-SA 3.0 (non-commercial, share-alike), inherited here. TO CONFIRM with the GaussianWorld authors for the splat fits.
Terms	https://www.cvlibs.net/datasets/kitti-360/
Homepage	https://huggingface.co/datasets/GaussianWorld/kitti_360_subset_chunk_gs_preprocessed
Cite	Liao Y., Xie J., Geiger A. KITTI-360: A Novel Dataset and Benchmarks for Urban Scene Understanding in 2D and 3D. TPAMI 2022.
Samples grouped by	chunk

Roles (what each sample contains)

Role	Type	Frame	Units	Labels / layout	Pattern or file
points	points	world	m	layout: (N,3) x, y, z	test/{chunk}/pc_coord.npy Accumulated LiDAR points of the chunk (KITTI-360 world frame, metres)
point_segments	attribute	from the file header	from the file header	layout: (N)	test/{chunk}/pc_segment.npy One segment id per LiDAR point (meaning of the ids TO CONFIRM)
gaussian_centers	points	world	m	layout: (N,3) x, y, z	test/{chunk}/coord.npy Centres of the fitted Gaussians (same frame as the points)
gaussian_color	attribute	from the file header	from the file header	layout: (N,3) r, g, b	test/{chunk}/color.npy RGB per Gaussian (range 0-1 or 0-255 TO CONFIRM)
gaussian_opacity	attribute	from the file header	from the file header	layout: (N,1)	test/{chunk}/opacity.npy
gaussian_scale	attribute	from the file header	from the file header	layout: (N,3)	test/{chunk}/scale.npy
gaussian_rotation	attribute	from the file header	from the file header	layout: (N,4)	test/{chunk}/quat.npy Unit quaternion per Gaussian

Standardized layout

Files are fetched on first access into the dataset cache and stay in their original format; ds.export(folder) writes every sample as standard files (PLY for points and meshes, MetaImage for volumes and label maps) with a .pardesline.json sidecar carrying frame, units, labels, licence and provenance, plus a manifest for the folder. Label maps become exact surfaces with to_mesh(label=k), DICOM series become one volume, raw arrays are decoded through the layout above.

ds = pl.load("hf_kitti360_gs_subset").take(2)
ds.export("./standardized")        # PLY / MetaImage + sidecars + manifest.json

Algorithms and models that run on it

• points roles: points.align(moving), points.clean(), points.to_mesh(), points.segment(eps, min_points) ; catalog modules: point_cloud_processing, mesh_processing, voxel_to_mesh, sdf, point_cloud_reconstruction, registration, deformable_registration, point_cloud_segmentation, stereo_3d_reconstruction, lidar_scene_detection, lidar_camera_fusion, features_extraction, collision_detection, pyvista_mesh_processing, pyvista_pointcloud_processing, pyvista_volume_processing, pyvista_analysis

Related

• Datasets catalog · How to download a medical imaging dataset in Python · Add a dataset
• Tags: lidar, points, splat, driving, kitti, hf, non-commercial

_{Generated from src/pardesline/data/manifests/hf_kitti360_gs_subset.json on 2026-10-14.}