Abstract: Highlights•We construct a new low-light SOD dataset, which consists of 3263 pairs of images.•We propose a novel low-light SOD model that integrates self-enhancement of saliency.•Our model performs competitively with fully supervised methods.
Loading