/explainable-cnn

Towards Visual Explanations for Convolutional Neural Networks via Input Resampling

Primary LanguagePython

explainable-cnn

Code to accompany the paper "Towards Visual Explanations for Convolutional Neural Networks via Input Resampling", presented at ICML 2017 Workshop on Visualization in Deep Learning.

Please contact blengeri [@] cs.cmu.edu with questions.

Citing

Please cite the following paper:

@article{lengerich2017visual,
  title={Visual Explanations for Convolutional Neural Networks via Input Resampling},
  author={Lengerich, Benjamin J and Konam, Sandeep and Xing, Eric P and Rosenthal, Stephanie and Veloso, Manuela},
  journal={arXiv preprint arXiv:1707.09641},
  year={2017}
}

Usage

python run_deconv.py

Dependencies

Used Keras with Theano as backend

  • Keras: (1.1.0)
  • Theano: (0.9.0dev2)
  • Pillow: (3.3.1)
  • h5py