Reflectance Sharing: Predicting Appearance from a Sparse Set of Images of a Known Shape