Improving Smooth GradCAM++ with Gradient Weighting Techniques

Convolutional Neural Networks (CNNs) have achieved unprecedented breakthroughs in a variety of computer vision tasks. Their black box nature however presents challenges in field such as health and security. As an effort to developing explainable deep learning models, several methods have been proposed to make CNNs more interpretable and trustworthy. We improve up one such method, SmoothGradCAM++ producing quantitavely better attribution maps. We test our approach on the validation set of the Imagenette,