We apply non-parameteric bayesian techniques to address a novel multi-class video co-segmentation challenge. Our method takes multiple videos as input and exploits commonalities between them in order to extract objects. Our formulation is based on distant-dependent Chinese Restaurant Processes.