3D face reconstruction is a versatile technology applied in various contexts. Deep learning methods, in particular, prove highly valuable in this domain due to their ability to generate high-quality 3D models. Nevertheless, most of these techniques require unobstructed and clear facial images as input. In response, we’ve developed a 3D face reconstruction system capable of performing effectively even when faced with obstructed views. Taking inspiration from generative face image inpainting, we’ve introduced a comprehensive method for crafting facial representations guided by their outlines. Our network comprises two components: one dedicated to image enhancement by removing obstructions and restoring missing facial features, while the other utilizes weak supervision to craft exceptionally detailed 3D models. Through extensive experimentation on standard 3D face reconstruction tasks, we’ve demonstrated the superiority of our method compared to existing ones that often fall short. Our results, based on experiments conducted with LFW databases, affirm the effectiveness of our approach.