Description

This repository implements the explainability techniques described in this paper: Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps.

Usage

This repo trains a ConvNet on the CIFAR10 dataset and then applies the visualization techniques of the paper to understand what the ConvNet does.

Training

To train the model, run:

python -m src.train

Explainability

To perform the explainability techniques, run:

python -m src.explain

Example output of Class Model Visualization:

Example output of Saliency Map: