We designed SymPaD framework, a model-driven visual dictionary construction and description method, with new shape models and quantized shape library. We demonstrate that, with this new design, the most current model-driven dictionary construction method is outperformed with even smaller dictionary in object recognition and image retrieval tasks.