Nate1874/image_explain

Explaining Image Classifier

The code for our TPAMI 2020 paper: Interpreting Image Classifiers by Generating Discrete Masks [paper link]

How to use our code

First, download the ImageNet dataset and set corresponding paths.

In configure.py, the configurations can be modified.

The build_imagenet_data.py and data_input.py files provide interface for data prepropossing and data loading.

The network.py defines the architectures of our model, including the G and D.

The model.py file include model training and testing.

In visual folder, we provide code to perform several comparing methods, such as gradients, CAM, guidedBP and IG.

Reference

@article{yuan2020interpreting,
  title={Interpreting image classifiers by generating discrete masks},
  author={Yuan, Hao and Cai, Lei and Hu, Xia and Wang, Jie and Ji, Shuiwang},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2020},
  publisher={IEEE}
}