HayCAM: A Novel Visual Explanation for Deep Convolutional Neural Networks

Explaining the decision mechanism of Deep Convolutional Neural Networks (CNNs) is a new and challenging area because “Black Box” nature CNN's. Class Activation Mapping (CAM) as visual explainable method used to highlight important regions input images by using classification gradients. The lack current methods use all filters in last convolutional layer which causes scattered unfocused activation mapping. HayCAM novel visualization provides better mapping therefore localization dimension reduction. It has been shown with mask detection case that are fed into CNN model bounding boxes drawn over generated maps (i.e. weakly-supervised object detection) three different CAM methods. IoU values obtained 0.1922 for GradCAM, 0.2472 GradCAM++, 0.3386 EigenCAM, 0.3487 proposed HayCAM. results show achieves best