The repository contains the implementation of DiffusionNOCS inference.
The paper and github pages are available:
- arXiv: https://arxiv.org/abs/2402.12647
- Github Pages: https://woven-planet.github.io/DiffusionNOCS/
You can try the inference of DiffusionNOCS by clicking the link below and following instructions in the colab noteboook.
git clone https://github.com/woven-planet/DiffusionNOCS.git
cd DiffusionNOCS
./download_weights.sh
python3 -m venv venv
source venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt
pip install -e .
ipython kernel install --user --name=venvIf you'd like to use your specific version, please modify requirements.txt based on your preference.
Run the following code:
jupyter notebook notebooks/diffusion_nocs.ipynbClick Kernel tab, then change kernel to venv.
Run the following code:
python3 scripts/inference.py --category-name "bottle"How to create the dataset of Generalization Benchmark
This work has been done at Woven by Toyota, Inc, and Toyota Research Institute.
Takuya Ikeda, Tianyi Ko, Robert Lee and Koichi Nishiwaki are with the Woven by Toyota, Inc.
Sergey Zakharov, Muhammad Zubair Irshad, Katherine Liu and Rares Ambrus are with the Toyota Research Institute.
The implementation was supported by Yuki Igarashi in Woven by Toyota, Inc. We'd like to express our deep gratitude to him.
@article{ikeda2024diffusionnocs,
title={DiffusionNOCS: Managing Symmetry and Uncertainty in Sim2Real Multi-Modal Category-level Pose Estimation},
author={Ikeda, Takuya and Zakharov, Sergey and Ko, Tianyi and Irshad, Muhammad Zubair and Lee, Robert and Liu, Katherine and Ambrus, Rares and Nishiwaki, Koichi},
journal={arXiv preprint arXiv:2402.12647},
year={2024}
}