A simple fusion framework that combines existing RGB-produced saliency with new depth-induced saliency and a specialized multi-stage RGBD model is proposed which takes account of both depth and appearance cues derived from low-level feature contrast, mid-level region grouping and high-level priors enhancement.