CADA-VAE

Original PyTorch implementation of "Generalized Zero-and Few-Shot Learning via Aligned Variational Autoencoders" (CVPR 2019).

Paper: https://arxiv.org/pdf/1812.01784.pdf

Requirements

The code was implemented using Python 3.5.6 and the following packages:

torch==0.4.1
numpy==1.14.3
scipy==1.1.0
scikit_learn==0.20.3
networkx==1.11

Using Python 2 is not recommended.

Data

Download the following folder https://www.dropbox.com/sh/btoc495ytfbnbat/AAAaurkoKnnk0uV-swgF-gdSa?dl=0 and put it in this repository. Next to the folder "model", there should be a folder "data".

Experiments

To run the experiments from the paper, navigate to the model folder and execute the following:

python single_experiment.py --dataset CUB --num_shots 0 --generalized True

The choices for the input arguments are:

datasets: CUB, SUN, AWA1, AWA2
num_shots: any number 
generalized: True, False

More hyperparameters can be adjusted in the file single_experiment.py directly. The results vary by 1-2% between identical runs.

Citation

If you use this work please cite

@inproceedings{schonfeld2019generalized,
  title={Generalized zero-and few-shot learning via aligned variational autoencoders},
  author={Schonfeld, Edgar and Ebrahimi, Sayna and Sinha, Samarth and Darrell, Trevor and Akata, Zeynep},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={8247--8255},
  year={2019}
}

Contact

For questions or help, feel welcome to write an email to [email protected]