Unsupervised identification of useful visual landmarks using multiple segmentations and top-down feedback