Abstract In this section we will present some advances in the visualization and understanding of convolutional networks. We will discuss how the different activation elements in a network correspond to specific image patches of an input image. We will also discuss about how textures in images are coded in a trained network. And finally, we will see how using these two visualization ideas result in the creation of artistic images that combine content and style in a very appealing form.
Class:
Data Science 2: Advanced Topics in Data Science