You can use any image as source, but to create annotations you have to do that yourself currently. Using simple segmentation libraries (or clustering) can do a good job for certain images, or look at better solutions for semantic segmentation: http://gitxiv.com/search/?q=segmentation