Skip to content

Reference implementation of saliency map techniques in Python 3 / Numpy

License

Notifications You must be signed in to change notification settings

andrewschreiber/numpy-saliency

Repository files navigation

numpy-saliency

Reference implementation of saliency map (aka attribution map) techniques for deep learning in Python 3. Uses MNIST and LeNet 5. Tested with 3.6 & 3.7.

Usage

Run

git clone https://github.com/andrewschreiber/numpy-saliency.git
cd numpy-saliency
# Activate your Python 3 environment
pip install -r requirements.txt
python main.py

You can train the model, test the model (using pretrained weights), and/or generate a saliency map via uncommenting the code.

Techniques

Implemented

Vanilla Gradients

Upcoming

Integrated Gradients

Adversarial Saliency maps

Inspiration

https://github.com/utkuozbulak/pytorch-cnn-visualizations

https://github.com/gary30404

About

Reference implementation of saliency map techniques in Python 3 / Numpy

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages