This repository contains the reference code for the paper "FREE: Feature Refinement for Generalized Zero-Shot Learning" accepted to ICCV 2021. [arXiv][Paper]

Datasets can be download from Xian et al. (CVPR2017) and take them into dir data.

The code implementation of FREE mainly based on PyTorch. All of our experiments run in Python 3.8.8.

Before running commands, you can set the hyperparameters in config.py. Please run the following commands and testing FREE on different datasets:

$ python ./image-scripts/run-cub.py       #CUB
$ python ./image-scripts/run-sun.py       #SUN
$ python ./image-scripts/run-flo.py       #FLO
$ python ./image-scripts/run-awa1.py      #AWA1
$ python ./image-scripts/run-awa2.py      #AWA2

Note: All of above results are run on a server with one GPU (Nvidia 1080Ti).

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

@InProceedings{Chen_2021_ICCV,
    author    = {Chen, Shiming and Wang, Wenjie and Xia, Beihao and Peng, Qinmu and You, Xinge and Zheng, Feng and Shao, Ling},
    title     = {FREE: Feature Refinement for Generalized Zero-Shot Learning},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    year      = {2021},
    pages     = {122-131}
}

We thank the following repos providing helpful components in our work.

1. TF-VAEGAN
2. cycle-CLSWGAN