wlj961012 / VAE-cFlow-ZSL

Implementation for "Generalized Zero-Shot Learning via VAE-Conditioned Generative Flow" （https://arxiv.org/abs/2009.00303）

VAE-cFlow-ZSL

Official Implementation for "Generalized Zero-Shot Learning via VAE-Conditioned Generative Flow"

Overview

Code Structure

├── README.md
├── train_vaepriorflow.py 
├── dataset_GBU.py        
├── classifier.py         
├── scripts/      
├── vaeflow/          
├── pics
├── data
│   ├── AWA1
│   ├── AWA2    
│   ├── CUB
│   └── SUN

How to use

1. Download data from Google Cloud

2. Set up conda environment

   conda create -n ZSL python=3.6
   conda install -c pytorch pytorch torchvision cudatoolkit=10.0
   

3. Train

   cd scripts/
   bash train_cub.sh
   

Cite

If you use this code base in your work, please cite

@article{gu2020generalized,
  title={Generalized Zero-Shot Learning via VAE-Conditioned Generative Flow},
  author={Gu, Yu-Chao and Zhang, Le and Liu, Yun and Lu, Shao-Ping and Cheng, Ming-Ming},
  journal={arXiv preprint arXiv:2009.00303},
  year={2020}
}