Auto-GAS: Automated Proxy Discovery for Training-free Generative Architecture Search

This is the code for the paper: Auto-GAS: Automated Proxy Discovery for Training-free Generative Architecture Search.

• Python 3.6
• PyTorch 1.0.0
• CUDA 9.0
• NVIDIA GPU + CUDA CuDNN

Download the dataset.

Train the search space.

python train.py --dataset mnist --data_path data --save_path save

Evaluate the search space.

python evaluate.py --dataset mnist --data_path data --save_path save

This code is based on the following projects.

• EAGAN
• NASLib
• TransBench

If you find Auto-GAS useful in your research, please consider citing the following paper:

@inproceedings{li2024auto,
  title={Auto-gas: Automated proxy discovery for training-free generative architecture search},
  author={Li, Lujun and Sun, Haosen and Li, Shiwen and Dong, Peijie and Luo, Wenhan and Xue, Wei and Liu, Qifeng and Guo, Yike},
  year={2024},
  organization={ECCV}
}

This project is licensed under the MIT License.