Deep edge map guided depth super resolution

Accurate edge reconstruction is critical for depth map super resolution (SR). Therefore, many traditional SR methods utilize maps to guide SR. However, it difficult predict accurate from low (LR) maps. In this paper, we propose a deep guided method, which includes an prediction subnetwork and subnetwork. The takes advantage of the hierarchical representation color images produce maps, promote performance disentangling cascaded network progressively upsample result, where every level made up weight sharing module adaptive module. extracts general features in different levels, while transfers specific adapt degraded inputs. Quantitative qualitative evaluations on various datasets with magnification factors demonstrate effectiveness promising proposed method. addition, construct benchmark dataset captured by Kinect-v2 facilitate research real-world