/AWGIM

Code for paper "Attentive Weights Generation for Few Shot Learning via Information Maximization"

Primary LanguagePython

Attentive Weights Generation for Few Shot Learning via Information Maximization

Published at CVPR 2020

By Yiluan Guo, Ngai-Man Cheung

Paper Link

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.

Citation

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}
}