Self-supervised Amodal Video Object Segmentation

SaV os also learns the spatiotemporal prior, which is also useful for the amodal task and could generalize to unseen types.