Deep Coral Tf

Deep Coral is attributed as a solution to the problem faced by convolutional neural networks to generalize well across changes in input distributions. Domain adaptation algorithms have been proposed to compensate for the degradation in performance due to domain shift. I have attempted to extend CORAL to learn a non-linear transformation that aligns correlations of the activations of the last fully connected layer in deep convolutional neural networks, hence the phrase Deep CORAL. Thus the network is trained on images of one domain like amazon and tested on images of other domain like webcam. The accuracy achieved on Office 31 dataset(comprising of 31 categories) is about 58%.