Skip to content

UniOcc, CARLA 10 Hz mini split: LiDAR sweeps, camera views and semantic occupancy grids

pl.load("hf_uniocc_carla_mini") loads this dataset. The smallest split of UniOcc (unified occupancy benchmark over nuScenes, Waymo, CARLA, OpenCOOD). Two CARLA scenes at 10 Hz; per time step a surround LiDAR sweep (.ply), four camera images, a bird's-eye view and an .npz with the 200x200x16 semantic occupancy grid, camera/LiDAR masks, voxel flow and the ego-to-world 4x4 transform. The other splits are folders of the same repo: pl.load('hf://tasl-lab/uniocc/NuScenes-via-Occ3D-2Hz-mini').

import pardesline as pl

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

Provenance and licence
Source Hugging Face Hub repo tasl-lab/uniocc, paths Carla-10Hz-mini
Scheme hf
Size of the whole source 4.9 GB
Licence MIT
Commercial use allowed by the licence
Redistribution allowed
Licence note MIT on the Hub card (checked 2026-10-13). The nuScenes- and Waymo-derived splits of the same repo inherit the upstream terms of those datasets (non-commercial for nuScenes) - this manifest covers the CARLA split only.
Terms https://huggingface.co/datasets/tasl-lab/uniocc
Homepage https://uniocc.github.io/
Cite Wang Y., Huang X., Sun X., Yan M., Xing S., Tu Z., Li J. UniOcc: A Unified Benchmark for Occupancy Forecasting and Prediction in Autonomous Driving. arXiv:2503.24381, 2025.
Samples grouped by scene

Roles (what each sample contains)

Role Type Frame Units Labels / layout Pattern or file
lidar points sensor m Carla-10Hz-mini/{scene}/{frame}_surround_lidar.ply
Surround LiDAR sweep in the ego-vehicle frame
occupancy labels from the file header from the file header layout: (200,200,16) Carla-10Hz-mini/{scene}/{frame}.npz
Semantic occupancy grid 200x200x16 (L, W, H), class ids per the UniOcc repo; voxel size and grid origin are not stated on the card, so the grid is loaded with unit spacing (TO CONFIRM: 0.4 m voxels in the Occ3D convention)
ego_to_world attribute from the file header from the file header layout: (4,4) Carla-10Hz-mini/{scene}/{frame}.npz
4x4 transform from the ego vehicle to the world frame (same .npz)
camera_0 image from the file header from the file header Carla-10Hz-mini/{scene}/{frame}_camera_0.png
camera_1 image from the file header from the file header Carla-10Hz-mini/{scene}/{frame}_camera_1.png
camera_2 image from the file header from the file header Carla-10Hz-mini/{scene}/{frame}_camera_2.png
camera_3 image from the file header from the file header Carla-10Hz-mini/{scene}/{frame}_camera_3.png
bev image from the file header from the file header Carla-10Hz-mini/{scene}/{frame}_bev.png
Bird's-eye-view rendering of the frame

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_uniocc_carla_mini").take(2)
ds.export("./standardized")        # PLY / MetaImage + sidecars + manifest.json

Algorithms and models that run on it

Generated from src/pardesline/data/manifests/hf_uniocc_carla_mini.json on 2026-10-14.