We propose a general multi-class visual recognition model, termed theClassifier Graph, which aims to generalize and integrate ideas from many oftoday's successful hierarchical recognition approaches. Our graph-based modelhas the advantage of enabling rich interactions between classes from differentlevels of interpretation and abstraction. The proposed multi-class system isefficiently learned using step by step updates. The structure consists ofsimple logistic linear layers with inputs from features that are automaticallyselected from a large pool. Each newly learned classifier becomes a potentialnew feature. Thus, our feature pool can consist both of initial manuallydesigned features as well as learned classifiers from previous steps (graphnodes), each copied many times at different scales and locations. In thismanner we can learn and grow both a deep, complex graph of classifiers and arich pool of features at different levels of abstraction and interpretation.Our proposed graph of classifiers becomes a multi-class system with a recursivestructure, suitable for deep detection and recognition of several classessimultaneously.
展开▼
