AffordGen: Generating Diverse Demonstrations for Generalizable Object Manipulation with Affordance Correspondence (CVPR 2026)

Install

conda create -n affordgen python=3.11
pip install -r requirements.txt

# Install affordgen
pip install -e affordgen

# Install sim_tasks
pip install -e sim_tasks

Demo Code

Checkout affordgen/demo.ipynb or affordgen/demo.py for demonstrating on teapot task.

Complete AffordGen generation scheme follows a 3-step prepration as shown in demo codes:

1. Camera parameters prepration
2. Source demo preprocessing
3. New demo configuration

3D Assets for Generation

The generated meshes used by AffordGen can be created using any text2mesh generation model. We will release the mesh dataset we previously used for generation tasks ASAP.

Cite

@article{zhang2026affordgen,
  title={AffordGen: Generating Diverse Demonstrations for Generalizable Object Manipulation with Afford Correspondence},
  author={Zhang, Jiawei and Hu, Kaizhe and Huang, Yingqian and Ju, Yuanchen and Xue, Zhengrong and Xu, Huazhe},
  journal={arXiv preprint arXiv:2604.10579},
  year={2026}
}