vision-cair / cizslv2

CIZSL++: Creativity Inspired Generative Zero-Shot Learning. T-PAMI under review.

CIZSLv2

CIZSL++: Creativity Inspired Generative Zero-shot Learning, Mohamed Elhoseiny, Kai Yi, Mohamed Elfeki, Arxiv, 2020

Requirements

Python 3.6

Pytorch 1.6

sklearn, scipy, matplotlib, numpy, random, copy

Processed Feature Data

You can download the text-based dataset at dataset CUBird and NABird. For attribute-based data, you can access to here.

Please put the uncompressed data to the folder "data".

Reproduce CIZSLv2 Best Model

python train_cizslv2.py --dataset 'CUB' --splitmode 'easy' --creativity_weight 1  --exp_name 'cizslv2'              
python train_cizslv2.py --dataset 'CUB' --splitmode 'hard' --creativity_weight 0.1  --exp_name 'cizslv2'                
python train_cizslv2.py --dataset 'NAB' --splitmode 'easy' --creativity_weight 0.001  --exp_name 'cizslv2'              
python train_cizslv2.py --dataset 'NAB' --splitmode 'hard' --creativity_weight 1  --exp_name 'cizslv2'

Reference

• Yizhe Zhu, Mohamed Elhoseiny, Bingchen Liu, Xi Peng and Ahmed Elgammal "A Generative Adversarial Approach for Zero-Shot Learning from Noisy Texts", CVPR, 2018
• Mohamed Elhoseiny, Mohamed Elfeki, Creativity Inspired Zero Shot Learning, Thirty-sixth International Conference on Computer Vision (ICCV), 2019

If you find this code is useful, please cite:

@article{elhoseiny2021cizsl++,
  title={CIZSL++: Creativity Inspired Generative Zero-Shot Learning},
  author={Elhoseiny, Mohamed and Yi, Kai and Elfeki, Mohamed},
  journal={arXiv preprint arXiv:2101.00173},
  year={2021}
}