wxyxixixi/artic-o-data

Artic-O — evaluation data

Test split and evaluation metadata for Artic-O: End-to-End Articulated Object Reconstruction via Latent Geometry Learning, SIGGRAPH Asia 2026 Conference Papers.

Project page · arXiv · Code

Derived from PartNet-Mobility. 312 test samples across 8 categories, each an (object, joint) pair rendered at two articulation states.

Contents

Path	Size	What
artic_o_test/	18 GB	the test split: 6 RGB views (3 at s0, 3 at s1) + 30k-point GT cloud + joint parameters per sample
val_views.json	45 KB	deterministic per-sample view picks — required for reproducible numbers
larm_train_test_list.json	350 KB	train/test split definition and object→category map

Important: this is save_to_disk format, not load_dataset format

artic_o_test/ is an Arrow directory written by Dataset.save_to_disk, so it must be downloaded and opened with load_from_disk. load_dataset will not work, and the dataset viewer is disabled for this reason.

from datasets import load_from_disk
ds = load_from_disk("./data/artic_o_test")["test"]   # 312 rows

Usage

hf download wxyxixixi/artic-o-data --repo-type dataset --local-dir ./data/

which produces the layout the code expects:

data/
  artic_o_test/{dataset_dict.json, test/}
  val_views.json
  larm_train_test_list.json

Code: https://github.com/Wxyxixixi/Artic-O · Checkpoint: wxyxixixi/artic-o

val_views.json is not optional. Without it the loader picks views at random per run, which adds roughly 3pp of variance to the articulation metrics and makes the numbers non-reproducible.

Citation

@article{wang2026artic,
  title={Artic-O: End-to-End Articulated Object Reconstruction via Latent Geometry Learning},
  author={Wang, Xuyang and Li, Zhenyu and Ding, Jian and Slim, Habib and Wonka, Peter and Li, Hongdong and Elhoseiny, Mohamed},
  journal={arXiv preprint arXiv:2606.21938},
  year={2026}
}

License

This license covers the data artifacts only. The code at https://github.com/Wxyxixixi/Artic-O is MIT licensed, and the paper is published by ACM under CC BY 4.0 — three separate licenses, none extending to the others.

Non-commercial research use only. Derived from PartNet-Mobility and remains subject to the PartNet-Mobility terms of use; obtain the original dataset from SAPIEN directly for any use beyond reproducing this paper's evaluation.