Abstract

We propose a fine-grained recognition system that incorporates part localization, alignment, and classification in one deep neural network. This is a nontrivial process, as the input to the classification module should be functions thatenable back-propagation in constructing the solver. Ourmajor contribution is to propose a valve linkage function(VLF) for back-propagation chaining and form our deep lo-calization, alignment and classification (LAC) system. TheVLF can adaptively compromise the errors of classificationand alignment when training the LAC model. It in turn help-s update localization. The performance on fine-grained object data bears out the effectiveness of our LAC system.