Structure Flow-Guided Network for Real Depth Super-resolution

Real depth super-resolution (DSR), unlike synthetic settings, is a challenging task due to the structural distortion and edge noise caused by natural degradation in real-world low-resolution (LR) maps. These defeats result significant structure inconsistency between map RGB guidance, which potentially confuses RGB-structure guidance thereby degrades DSR quality. In this paper, we propose novel flow-guided framework, where cross-modality flow learned guide information transferring for precise upsampling. Specifically, our framework consists of upsampling network (CFUNet) flow-enhanced pyramid attention (PEANet). CFUNet contains trilateral self-attention module combining both geometric semantic correlations reliable learning. Then, maps are combined with grid-sampling mechanism coarse high-resolution (HR) prediction. PEANet targets at integrating as into hierarchically learn edge-focused feature refinement. Extensive experiments on real datasets verify that approach achieves excellent performance compared state-of-the-art methods. Our code available at: https://github.com/Yuanjiayii/DSR-SFG.