We present a novel approach to unsupervised learning for video object segmentation (VOS). In contrast to previous methods, our approach learns dense feature representations directly in a fully-convolutional regime. We rely on uniform grid sampling to extract a set of anchors and train our model to disambiguate between them on the inter- and intra-video levels. A naive scheme to train such a model results in a degenerate solution. However, we prevent it with simple regularisation accommodating eq…