This project provides a strong Baseline with WRN28-10 and ResNet-10 backbone for the following Few-Shot Learning methods:
This also includes implementation of our NeurIPS 2020 paper Interventional Few-Shot Learning, which proposes IFSL classifier based on intervention P(Y|do(X)) to remove the confounding bias from pre-trained knowledge. Our IFSL classifier is generally applicable to all fine-tuning and meta-learning method, easy to plug in and involves no additional training steps.
The codes are organized into four folders according to methods. The folder MAML_MN_FT contains baseline and IFSL for fine-tuning, Matching Networks and MAML.
Recommended version:
After downloading the weights and datasets, you can follow the instructions in each folder to modify the code and finish preparation.
Apologize in advance for dirty code, which I will clean up gradually.
The implementation is based on the following repositories (for correctness of baseline, most of our code is based on the official released code).
If you find our work or the code useful, please consider cite our paper using:
@inproceedings{yue2020interventional,
title={Interventional Few-Shot Learning},
author={Yue, Zhongqi and Zhang, Hanwang and Sun, Qianru and Hua, Xian-Sheng},
booktitle= {NeurIPS},
year={2020}
}