本项目由AI Studio对应项目迁移而来,请访问AI Studio获取对应项目运行方式。具体技术细节可参考notebook说明。您可以通过如下方式快捷地运行本项目:
首先下载数据集,并放到如下文件夹下:
- data
- cifar-100-python.tar.gz
- Track1_final_archs.json
通过如下命令来训练
python -m supernet.scripts.train --n 18 --kd --sandwich
通过如下命令来测试
python -m supernet.scripts.evaluate --path path/to/saved/model --output path/to/output
如果您发现相关代码和项目对您有用,请引用我们的文章:
@inproceedings{guan2021oneshot,
title={One-Shot Neural Channel Search: What Works and What’s Next},
author={Chaoyu Guan and Yijian Qin and Zhikun Wei and Zeyang Zhang and Zizhao Zhang and Xin Wang and Wenwu Zhu},
booktitle={CVPR 2021 Workshop on Neural Architecture Search: 1st lightweight NAS challenge and moving beyond},
year={2021}
}