The Mean-Shift procedure was introduced to image processing and used for feature extraction and segmentation of images and video. This procedure works as a gradient descent in 'feature-space' to find maxima of an estimation of a probability density function. In this talk I will discuss novel algorithms which adapt the Mean-Shift procedure to work on volumetric and embedded boundary meshes, structured or unstructured with both scalar and vector attributes. The attributes, which are either given (temperature, pressure etc.) or extracted (curvature, geodesic centricity) are combined to create the 'feature-space', and the algorithm uses some local approximations of this space to constrain and accelerate the process. The mean shift algorithm can be utilized for data-exploration in visualization, for feature-extraction, for partitioning, and more.