• Overview
• Installation
• Contributing
• Citation
• License
• TODO

XNAS is an effective, modular and flexible Neural Architecture Search (NAS) repository, which aims to provide a common framework and baselines for the NAS community. It is originally designed to decouple the search space, search algorithm and performance evaluation strategy to achieve freely combinable NAS.

This project is now supported by PengCheng Lab.

---

Overview | Installation | Contributing | Citation | License

For more information and API usages, please refer to our Documentation.

Overview

Based on a common division of NAS, the project is organized by search space, search algorithm, and evaluation strategy. The project currently supports the content shown below. Each row of the table represents a search algorithm and each column represents the search space.

	DARTS	NAS-Bench-201	SPOS	OFA	NAS-Bench-101	NAS-Bench-1Shot1
DARTS	C10 C100	C10 C100 ImageNet16
PDARTS	C10 ImageNet
PCDARTS
SNG
ASNG
MDENAS
DDPNAS
MIGONAS
GridSearch
DrNAS
RMINAS	C10 ImageNet	C10 C100 ImageNet
DropNAS
SPOS
OFA

We also provide the interpretation of papers and experimental records for each algorithm. For more information, please refer to the links in the "Docs" column.

Search Spaces	Docs	Official Links
DARTS		Github
MobileNetV3		-
NAS-Bench-101		GitHub
NAS-Bench-201		GitHub
NAS-Bench-1Shot1		GitHub
SPOS		GitHub
Search Algorithms	Docs	Official Links
DARTS		Github
PDARTS		Github
PCDARTS		Github
SNG		Github
ASNG		Github
MDENAS		Github
DDPNAS		Github
MIGONAS		Openi
GridSearch		-
DrNAS		Github
RMINAS		Openi
DropNAS		Github
SPOS		Github
OFA		Github
Evaluation Strategies	Docs	Official Links
NAS-Bench-101		GitHub
NAS-Bench-201		GitHub
NAS-Bench-301		GitHub
NAS-Bench-1Shot1		GitHub

We are gradually providing support for more settings.

Installation

To run XNAS, python>=3.7 and pytorch=1.9 are required. Other versions of PyTorch may also work well, but there are potential API differences that can cause warnings to be generated.

For detailed instructions, please refer to get_started.md and data_preparation.md in our docs.

Contributing

We welcome contributions to the library along with any potential issues or suggestions.

Please refer to Contributing.md in our docs for more information.

Citation

If you use this code in your own work, please use the following bibtex entries:

@inproceedings{zheng2022rminas,
  title={Neural Architecture Search with Representation Mutual Information},
  author={Xiawu Zheng, Xiang Fei, Lei Zhang, Chenglin Wu, Fei Chao, Jianzhuang Liu, Wei Zeng, Yonghong Tian, Rongrong Ji},
  journal={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2022}
}
@article{zheng2021migo,
  title={MIGO-NAS: Towards fast and generalizable neural architecture search},
  author={Zheng, Xiawu and Ji, Rongrong and Chen, Yuhang and Wang, Qiang and Zhang, Baochang and Chen, Jie and Ye, Qixiang and Huang, Feiyue and Tian, Yonghong},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2021},
  publisher={IEEE}
}
@inproceedings{zheng2020rethinking,
  title={Rethinking performance estimation in neural architecture search},
  author={Zheng, Xiawu and Ji, Rongrong and Wang, Qiang and Ye, Qixiang and Li, Zhenguo and Tian, Yonghong and Tian, Qi},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={11356--11365},
  year={2020}
}

License

This project is released under the MIT license.
XNAS also uses codes from these repos to build some modules:

• pycls
• pt.darts
• once-for-all

TODO

• 迁移OFA代码
• 补充101安装测试
• NAS-Bench-Macro
• 多显卡支持