EGANS

This is the codes of paper "EGANS: Evolutionary Generative Adversarial Network Search for Zero-Shot Learning" accepted to IEEE Transactions on Evolutionary Computation (TEC).

Requirements

The code implementation of EGANS mainly based on PyTorch. All of our experiments run and test in Python 3.10.6.

We use CLSWGAN as the baseline to verify our method and conduct experiments on four popular ZSL benchmarks: CUB, SUN, FLO and AWA2 following the data split of xlsa17.

Evolution Generator Architecture Search

Generator Architecture Searching Script

$ python clswgan_G_search.py

Evolution Discriminator Architecture Search

Discriminator Architecture Searching Script

$ python clswgan_D_search.py

Zero-shot prediction

Model retrain and final prediction

$ python clswgan_retrain.py

Citation

If this work is helpful for you, please cite our paper.

@ARTICLE{10225587,
  author={Chen, Shiming and Chen, Shuhuang and Hou, Wenjin and Ding, Weiping and You, Xinge},
  journal={IEEE Transactions on Evolutionary Computation}, 
  title={EGANS: Evolutionary Generative Adversarial Network Search for Zero-Shot Learning}, 
  year={2023},
  volume={},
  number={},
  pages={1-1},
  keywords={Computer architecture;Generators;Generative adversarial networks;Visualization;Training;Semantics;Optimization;Evolutionary neural architecture search;zero-shot learning;generative adversarial networks},
  doi={10.1109/TEVC.2023.3307245}}

Contact

If you have any questions, please drop email to gchenshiming@gmail.com or shuhuangchen@hust.edu.cn.

shiming-chen/EGANS