Matlab and Python Codes for the paper 'Zero-shot Learning via Recurrent Knowledge Transfer'.
You can download all data (image features, attribtues and word vectors of AwA, CUB and ImageNet) used in this paper from google drive. Then, put the data and code in the same fold (root path of the project).
Before running the code, you need to install two toolboxes, namely, Dimensionality Reduction toolbox (drtoolbox.tar.gz) and LeastR toolbox (SLEP_package_4.1.zip). First, download the two toolboxes from google drive. Then, unzip drtoolbox.tar.gz and SLEP_package_4.1.zip. Finally, add the two toolboxes into your Matlab path with subfolders.
For more support, you can visit drtoolbox and LeastR.
Now, you can run main.m.
The cross validation is implemented in RecKT_CV.m.
The algorithm (RecKT) is implemented in RecKT.m.
The Python code is a simple demo.
Put data into folder '../data/'
Run RKT.py
@inproceedings{zhao2019zero,
title={Zero-Shot Learning Via Recurrent Knowledge Transfer},
author={Zhao, Bo and Sun, Xinwei and Hong, Xiaopeng and Yao, Yuan and Wang, Yizhou},
booktitle={2019 IEEE Winter Conference on Applications of Computer Vision (WACV)},
pages={1308--1317},
year={2019},
organization={IEEE}
}
If you have any questions, feel free to contact me.
Email Address: bozhao at pku.edu.cn