Orthogonal Representations of Object Shape and Category in Deep Convolutional Neural Networks and Human Visual Cortex