Visual Boundary Knowledge Translation for Foreground Segmentation

When confronted with objects of unknown types in an image, humans can effortlessly and precisely tell their visual boundaries. This recognition mechanism underlying generalization capability seem to contrast state-of-the-art image segmentation networks that rely on large-scale category-aware annotated training samples. In this paper, we make attempt towards building models explicitly account for boundary knowledge, hope reduce the effort segmenting unseen categories. Specifically, investigate a new task termed as Boundary Knowledge Translation (BKT). Given set fully labeled categories, BKT aims translate knowledge learned from novel each which is provided only few To end, propose Segmentation Network (Trans-Net), comprises network two discriminators. The network, combined boundary-aware self-supervised mechanism, devised conduct foreground segmentation, while discriminators work together adversarial manner ensure accurate categories under light supervision. Exhaustive experiments demonstrate that, tens samples guidance, Trans-Net achieves close results par supervised methods.