Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

This paper addresses the visualisation of image classification models, learnt deep Convolutional Networks (ConvNets). We consider two visualisation, based on computing the gradient of the class score with respect to input image. The first one generates an image, which maximises the class Erhan et al. , 2009, thus visualising the notion of the class, captured a ConvNet. The second technique computes a class saliency map, specific to a image and class. We show that such maps can be employed for weakly object segmentation using classification ConvNets. Finally, we the connection between the gradient-based ConvNet visualisation and deconvolutional networks Zeiler et al. , 2013.