Defensive Few-Shot Learning. TPAMI 2023.

This is a PyTorch implementation of the DFSL paper.

Dependencies

If you don't have python 3 environment:

conda create -n DFSL python=3.8
conda activate DFSL

Then install the required packages:

pip install -r requirements.txt

Datasets

Caltech-UCSD Birds-200-2011, Standford Cars, Standford Dogs, miniImageNet and tieredImageNet are available at Google Drive and 百度网盘(提取码：yr1w).

Train & Test

1. We make the training and testing in a single script file.
2. Training Baseline with adversarial training and using DN4 for test:

     Python Baseline_AT_Test_DN4.py

3. Training DFSL and using DN4 for both training and test (A FGSM attacker is used.):

     Python DFSL_DN4_FGSM.py

4. Training DFSL and using DN4 for both training and test (A PGD attacker is used.):

     Python DFSL_DN4_PGD.py

Citation

@ARTICLE{9916072,
  author={Li, Wenbin and Wang, Lei and Zhang, Xingxing and Qi, Lei and Huo, Jing and Gao, Yang and Luo, Jiebo},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
  title={Defensive Few-Shot Learning}, 
  year={2023},
  volume={45},
  number={5},
  pages={5649-5667},
  doi={10.1109/TPAMI.2022.3213755}}