uCO3D: UnCommon Objects in 3D
- Provider
- Meta AI (FAIR)
- License
- CC-BY-4.0 (source)
- Size
- 4,189 files
- Stored at
- Hugging Face (https://huggingface.co/datasets/facebook/uco3d)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Computer Vision
- Source
- https://arxiv.org/abs/2501.07574
About uCO3D: UnCommon Objects in 3D
uCO3D (UnCommon Objects in 3D) is Meta AI's object-centric 3D dataset, the successor to CO3Dv2 (arXiv:2501.07574). It contains about 170,000 turn-table videos that circle real objects from all directions, covering roughly 1,000 categories of the LVIS taxonomy grouped into 50 super-categories. Unlike CO3Dv2 it releases the full original videos rather than extracted frames. Each video is annotated with object segmentation masks, camera poses, depth maps, three kinds of point clouds (dense, segmented and sparse), a 3D Gaussian Splat reconstruction, and a long and a short caption produced by a video-language model. Per sequence the files are rgb_video.mp4, mask_video.mkv, depth_maps.h5, point_cloud.ply, segmented_point_cloud.ply, sparse_point_cloud.ply and a gaussian_splats folder, indexed by metadata.sqlite and set_lists sqlite files. The authors report 19.27 TB in total (rgb_videos 7.59 TB, depth_maps 9.69 TB, gaussian_splats 1.18 TB). Measured 2026-11-14: the repository holds metadata.zip, examples.zip and thousands of part_ _NNNN.zip chunks of up to about 20 GB each, plus a vggt_omega_anno folder; the Hub reports about 28 TB stored. Use the uco3d Python package from GitHub to download, unzip and load sequences. Uses: novel view synthesis, 3D reconstruction, image-to-3D and text-to-3D generation, camera pose estimation. Released under CC BY 4.0; cite the uCO3D paper.
How to load uCO3D: UnCommon Objects in 3D in Python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="facebook/uco3d", filename="examples.zip", repo_type="dataset")
print(path)Data structure
Files
Example file: examples.zip
Analyze this dataset without downloading it
On Scigantic, uCO3D: UnCommon Objects in 3D opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Computer Vision datasets
- Textvqa
- YouTube UGC Dataset: User-Generated Content Video Clips with Mean Opinion Scores for Video Quality Assessment and Compression (Google)
- Nemotron VLM Dataset V 2
- PUG Animals
- Minimal Video Pairs
- ABU: Airport-Beach-Urban hyperspectral anomaly/target detection benchmark
- Artemis and Image Safari Agricultural Imaging Dataset
- Nemotron Personas India