Unified Mask Embedding and Correspondence Learning for Self-Supervised Video Segmentation
This work addresses the problem of reducing annotation costs for video object segmentation, making it more accessible, though it is incremental as it builds on prior self-supervised efforts.
The paper tackles self-supervised video object segmentation by developing a unified framework that learns cross-frame correspondence and object-level mask embedding from unlabeled videos, achieving state-of-the-art results on DAVIS17 and YouTube-VOS benchmarks and narrowing the gap with fully supervised methods.
The objective of this paper is self-supervised learning of video object segmentation. We develop a unified framework which simultaneously models cross-frame dense correspondence for locally discriminative feature learning and embeds object-level context for target-mask decoding. As a result, it is able to directly learn to perform mask-guided sequential segmentation from unlabeled videos, in contrast to previous efforts usually relying on an oblique solution - cheaply "copying" labels according to pixel-wise correlations. Concretely, our algorithm alternates between i) clustering video pixels for creating pseudo segmentation labels ex nihilo; and ii) utilizing the pseudo labels to learn mask encoding and decoding for VOS. Unsupervised correspondence learning is further incorporated into this self-taught, mask embedding scheme, so as to ensure the generic nature of the learnt representation and avoid cluster degeneracy. Our algorithm sets state-of-the-arts on two standard benchmarks (i.e., DAVIS17 and YouTube-VOS), narrowing the gap between self- and fully-supervised VOS, in terms of both performance and network architecture design.