A method that utilized Generative Adversarial Network (GAN) to interpret the black-box deep image classifier models by PyTorch.

Requirements

1. The project was implemented and tested in Python 3.5 and Pytorch 0.4. The higher versions should work after minor modification.
2. Other common modules like numpy, pandas and seaborn for visualization.
3. NVIDIA GPU and cuDNN are required to have fast speeds. For now, CUDA 8.0 with cuDNN 6.0.20 has been tested. The other versions should be working.

Datasets

Our proposed Chinese Character dataset is accessible on link

Implementation details

data preparation

build train/validation/test sets,

1-make_chinese1_list.py
2-make_letter_list.py
3-make_tiny_letter_list.py

training

4-train_capital_letter_resnet20.py
5-train_chinese1_resnet20.py
6-train_tiny_letter_resnet20.py

explanation generation

7-FARCE_capital_letter.py
8-FARCE_chinese.py
9-FARCE_mnist.py

Time and Space

All experiments were run on NVIDIA TITAN Xp