This repository contains the official code release for the Articulate3D paper, accepted at ICCV 2025.
📄 Paper: Articulate3D (ICCV 2025)
🏁 Challenge: Track 3 at OpenSUN3D Workshop, ICCV 2025
🛠 Code: Data loader available now, full method (USDNet) to be released by end of July 2025
Currently released:
SceneDataLoader: A Python class for loading and parsing Articulate3D annotations.
Coming soon (July 2025):
- Full implementation of USDNet, the method proposed in the paper.
Join the Articulate3D Challenge at the OpenSUN3D Workshop (ICCV 2025)!
We're hosting Track 3, which focuses on articulated scene understanding.
📍 Challenge details and submission portal: OpenSUN3D Challenge
SceneDataLoader is a Python iterator that loads Articulate3D annotations from its dataset directory.
Each scene is composed of:
<scene_id>_parts.json: part annotations and mesh face indices.<scene_id>_artic.json: articulation parameters (axis, origin, range, type).
from loader import SceneDataLoader
loader = SceneDataLoader("path/to/Articulate3D/")
for scene_id, scene_dict, face_mask in loader:
print(f"Scene: {scene_id}")
print(f"Articulated parts: {list(scene_dict.keys())}")
print(f"Face mask shape: {face_mask.shape}")If you find our work useful, please cite our paper.
@article{halacheva2024articulate3d,
title={Holistic Understanding of 3D Scenes as Universal Scene Description},
author={Anna-Maria Halacheva and Yang Miao and Jan-Nico Zaech and Xi Wang and Luc Van Gool and Danda Pani Paudel},
year={2024},
journal={arXiv preprint arXiv:2412.01398},
}