Parametric Optimization of Friction Stir Welding (FSW) of Dissimilar Aluminum Alloys with Newly Developed Tool

Abstract

Semantic Image Inpainting is a challenging and promising research problem where parts of an image are masked or corrupted and need to be filled based on the semantic and visual information available from the image. Image Inpainting can help revive ancient scriptures, paintings, and recover corrupted parts of images without the guidance of a subject matter expert. Existing computer vision techniques using neighborhood based gap filling provide unsatisfactory results. In this paper, a novel visual feature learning model driven by context-based pixel prediction is employed, which develops the missing part of the image by conditioning on the available data. We adopt a Generative model based approach to derive inference about the missing part of the image from the context of the entire image. A novel loss function based on the linear combination of generative loss and contextual loss provides promising results for semantic image inpainting. Experiments on datasets show that our method successfully predicts information in large missing regions and achieves pixel-level photorealism.