摘要
Completing a corrupted image by filling in correct structures and reasonable textures for a complex scene remains an elusive challenge. In case that a missing hole involves diverse semantic information, conventional two-stage approaches based on structural information often lead to unreliable structural prediction and ambiguous visual texture generation. To address the problem, we propose a SEmantic GUidance and Estimation Network (SeGuE-Net) that iteratively evaluates the uncertainty of inpainted visual contents based on pixel-wise semantic inference and optimize structural priors and inpainted contents alternatively. Specifically, SeGuE-Net utilizes semantic segmentation maps as guidance in each iteration of image inpainting, under which location-dependent inferences are re-estimated, and, accordingly, poorly-inferred regions are refined in subsequent iterations. Extensive experiments on real-world images demonstrate the superiority of our proposed method over state-of-the-art approaches in terms of clear boundaries and photo-realistic textures.