Tree-CNN: from generalization to specialization

Traditional convolutional neural networks (CNNs) classify all categories by a single network, which passes all kinds of samples through totally the same network flow. In fact, it is quite challengeable to distinguish schooner with ketch and chair by a single network. To address it, we propose a new image classification architecture composed of a cluster algorithm and the Tree-CNN. The cluster algorithm devotes to classifying similar fine categories into a coarse category. The Tree-CNN is comprised of a Trunk-CNN for coarse classification of all categories and Branch-CNNs to treat different groups of similar categories differently. Branch-CNNs are fine-tuning based on the Trunk-CNN, which extracts the special feature map of image and divides it into fine categories. But Branch-CNNs bring extra computation and are hard to train. To address it, we introduce adaptive algorithm to balance the heavy computation and accuracy. We have tested Tree-CNNs based on CaffeNet, VGG16, and GoogLeNet in Caltech101 and Caltech256 for image classification. Experiment results show the superiority of the proposed Tree-CNN.