Interpretable CNNs

The goal of the project is to develop a small toolkit that will be useful to analyze CNN model and gain more insights about how model is working.

Project structure

We have packaged a few visualisation algorithms into class that is able get different visualisation from VGG network. (ClassifierExplainer)

We have trained an example neural network on CelebA dataset to classify Smiling/Not Smiling people.

Example Visualisations (viz_demo.ipynb)

Image source	Image
Original image
Heatmap of activations
Smoothed gradient
Last layer filter patterns
Image after last layer filter
Image after 1st layer filter

Running locally

All visualizations are located in the vis_demo.ipynb notebook.

Before running check Virtualenv manual

#Create virtualenv (Run in project folder)
virtualenv --python=python3 .env

#Activate environment
source .env/bin/activate

#Install packages (with activated env)
pip install -r requirements.txt

Celeba dataset is used for training and evaluation.

To download dataset use:

kaggle datasets download -d jessicali9530/celeba-dataset 

References

Repository with visualisation algorithms