Published at CVPR 2020
By Yiluan Guo, Ngai-Man Cheung
The implementation is written in Python 3 and has been tested on tensorflow 1.12.0, Ubuntu 16.04.
Parts of the code are borrowed from LEO.
The feature embeddings for miniImageNet and tieredImageNet can be downloaded from https://github.com/deepmind/leo.
5-way 1-shot experiment on miniImageNet:
python main.py
The hyper-parameters can be tuned in main.py
and AWGIM is in model.py
.
Please cite our work if you find it useful in your research:
@inproceedings{guo2020awgim,
title = {Attentive Weights Generation for Few Shot Learning via Information Maximization},
author = {Yiluan Guo, Ngai-Man Cheung},
booktitle = {CVPR},
year = {2020}
}