SketchMobility / README.md
Arlo397's picture
SketchMobility dataset release
7ce886f
|
Raw
History Blame Contribute Delete
3.2 kB
---
language:
- en
pretty_name: SketchMobility
tags:
- 3D
- articulated-objects
- robotics
- urdf
- motion
size_categories:
- 1K<n<10K
viewer: false
license: other
extra_gated_heading: Request access to SketchMobility
extra_gated_prompt: >-
Please provide the information below and acknowledge the applicable
SketchMobility and upstream license terms described in LICENSE.md and
LICENSE_MAP.json. Access is granted automatically after submission.
extra_gated_fields:
Name: text
PI/Advisor: text
Affiliation: text
Purpose: text
Country: country
I have read and agree to the applicable licenses: checkbox
---
# SketchMobility
SketchMobility is a curated release of **4,956 articulated 3D objects** with
human-reviewed motion annotations, URDF descriptions, source-sensitive mesh
assets, and per-object normalization metadata.
This dataset accompanies
[Sketch2Arti: Sketch-based Articulation Modeling of CAD Objects](https://arxiv.org/abs/2604.25781).
## Dataset
| Source | Objects | License treatment |
| --- | ---: | --- |
| [Articraft](https://articraft3d.github.io/) (`Agentic`) | 205 | CC BY 4.0 |
| [Infinigen-Articulated](https://huggingface.co/datasets/princeton-vl/infinigen-articulated) | 726 | CC BY 4.0 |
| [PartNeXt](https://huggingface.co/datasets/AuWang/PartNeXt) | 2,177 | CC BY 4.0 |
| [Shape2Motion](https://github.com/wangxiaogang866/Shape2Motion) | 1,848 | GPL-3.0 |
| **Total** | **4,956** | **Mixed** |
After unpacking, each object has this layout:
```text
data/
{source}/
{category}/
{display_id}/
annotation.json
mobility.urdf
meshes/
```
`annotation.json` retains the original `upstreamId`, release ID, joints, mesh
links, and exact normalization transform. Mesh encodings remain
source-sensitive.
Sketch images are not included. The sketch-generation script will be released
with the Sketch2Arti system code.
## Download and unpack
The dataset follows the Artiverse release convention: verified `tar.gz` chunks
are stored under `dataset_chunks/`, and each object directory stays within one
chunk.
```bash
hf download Arlo397/SketchMobility \
--repo-type dataset \
--local-dir SketchMobility
cd SketchMobility
python pack_dataset_chunks.py unpack \
--chunks-dir dataset_chunks \
--overwrite \
--verify-sha256
```
The command reconstructs `data/`. Chunk sizes and SHA-256 checksums are listed
in `dataset_chunks/manifest.json`.
## License and attribution
SketchMobility is a mixed-license compilation. SketchMobility-original
curation, annotations, corrections, normalization metadata, release
identifiers, and documentation are CC BY 4.0 to the extent these rights are
held by the contributors. Upstream terms continue to apply.
See [LICENSE.md](LICENSE.md), [LICENSE_MAP.json](LICENSE_MAP.json), and
[ATTRIBUTION.md](ATTRIBUTION.md).
## Citation
```bibtex
@misc{yang2026sketch2arti,
title = {Sketch2Arti: Sketch-based Articulation Modeling of CAD Objects},
author = {Yang, Yi and Pan, Hao and Cui, Yijing and Sheffer, Alla and Li, Changjian},
year = {2026},
eprint = {2604.25781},
archivePrefix = {arXiv},
primaryClass = {cs.CV}
}
```