Unsupervised video object segmentation: An affinity and edge learning approach