Jianfeng Zhang 张健锋

Pygmalion: Bridging Reconstruction and Generation in Sparse Voxel-based 3D Modeling

Pygmalion: sparse voxel VAEs reconstruct well but generate poorly, as tiny diffusion errors tear holes in the mesh. By decoding continuous SDFs, Pygmalion closes this gap and improves steadily with DiT size and voxel resolution.

@inproceedings{luo2026pygmalion,
  title={{Pygmalion: Bridging Reconstruction and Generation in Sparse Voxel-based 3D Modeling}},
  author={Luo, Guan and Lin, Jing and Yi, Xuanyu and Liu, Jiahang and Zhang, Song-Hai and Zhang, Jianfeng},
  booktitle={Advances in Neural Information Processing Systems},
  year={2026}
}

ArtiAtlas: A Holistic Ecosystem for Diverse 3D Articulation Modeling

ArtiAtlas: a holistic ecosystem for diverse 3D articulation modeling, turning static 3D assets into articulated, functional objects.

@inproceedings{lu2026artiatlas,
  title={{ArtiAtlas: A Holistic Ecosystem for Diverse 3D Articulation Modeling}},
  author={Lu, Junzhe and Lin, Jing and Cao, Ziang and Wang, Haoqian and Liu, Ziwei and Zhang, Jianfeng},
  booktitle={ACM SIGGRAPH Asia Conference Papers},
  year={2026}
}

Seed3D 2.0: Advancing High-Fidelity Simulation-Ready 3D Content Generation

Seed3D 2.0: a new-generation 3D foundation model with a coarse-to-fine geometry pipeline and a unified PBR material model, extended to simulation-ready scene generation via layout planning, part-aware decomposition, and training-free articulation.

Seed3D 1.0: From Images to High-Fidelity Simulation-Ready 3D Assets

Seed3D 1.0: a 3D generative foundation model that generates simulation-ready assets from single images, enabling scalable physics-based world simulators!

Puppeteer: Rig and Animate Your 3D Models

What if static 3D models could be automatically rigged and animated? Puppeteer makes it possible, transforming any 3D object into animation-ready assets with stable motion!

@inproceedings{song2025puppeteer,
  title={{Puppeteer: Rig and Animate Your 3D Models}},
  author={Song, Chaoyue and Li, Xiu and Yang, Fan and Xu, Zhongcong and Wei, Jiacheng and Liu, Fayao and Feng, Jiashi and Lin, Guosheng and Zhang, Jianfeng},
  booktitle={Advances in Neural Information Processing Systems},
  year={2025}
}

MagicArticulate: Make Your 3D Models Articulation-Ready

MagicArticulate transforms static 3D models into articulation-ready assets with 1) Articulation-XL, a large-scale benchmark of 33K+ annotated models; 2) auto-regressive skeleton generation handling varying bone structures; and 3) functional diffusion for skinning with geodesic priors.

@inproceedings{song2025magicarticulate,
  title={{MagicArticulate: Make Your 3D Models Articulation-Ready}},
  author={Song, Chaoyue and Zhang, Jianfeng and Li, Xiu and Yang, Fan and Chen, Yiwen and Xu, Zhongcong and Liew, Jun Hao and Guo, Xiaoyang and Liu, Fayao and Feng, Jiashi and Lin, Guosheng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2025}
}

Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders

3D shape VAEs used to lose geometric details with uniform sampling, but Dora's sharp edge sampling preserves them while being 8× smaller.

@inproceedings{chen2025dora,
  title={{Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders}},
  author={Chen, Rui and Zhang, Jianfeng and Liang, Yixun and Luo, Guan and Li, Weiyu and Liu, Jiarui and Li, Xiu and Long, Xiaoxiao and Feng, Jiashi and Tan, Ping},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2025}
}

MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model

A breakthrough in human animation: MagicAnimate delivers the first high-fidelity, temporally consistent results via video diffusion model!

@inproceedings{xu2024magicanimate,
  title={{MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model}},
  author={Xu, Zhongcong and Zhang, Jianfeng and Liew, Jun Hao and Yan, Hanshu and Liu, Jiawei and Zhang, Chenxu and Feng, Jiashi and Shou, Zheng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2024}
}

AvatarGen: A 3D Generative Model for Animatable Human Avatars

Generative 3D humans that are both high-quality and animatable? AvatarGen achieves this through SMPL-guided canonical representation, trained only from 2D images.

@inproceedings{zhang2022avatargen,
  title={{AvatarGen: A 3D Generative Model for Animatable Human Avatars}},
  author={Zhang, Jianfeng and Jiang, Zihang and Yang, Dingdong and Xu, Hongyi and Shi, Yichun and Song, Guoxian and Xu, Zhongcong and Wang, Xinchao and Feng, Jiashi},
  booktitle={European Conference on Computer Vision Workshops},
  year={2022}
}

PoseAug: A Differentiable Pose Augmentation Framework for 3D Human Pose Estimation

3D pose estimators struggle with generalization? PoseAug solves this through differentiable, online pose augmentation that learns to generate diverse yet plausible training poses!

@inproceedings{gong2021poseaug,
  title={{PoseAug: A Differentiable Pose Augmentation Framework for 3D Human Pose Estimation}},
  author={Gong, Kehong and Zhang, Jianfeng and Feng, Jiashi},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2021}
}