Human-Aware Image Inpainting Using Neural Networks

Automatic masking and inpainting of people in images is a challenging problem in computer vision and image processing. The paper unveils the prime challenges in the process of object inpainting, including complex and diverse backgrounds. Current work also discusses the challenges in the process of masking. This manuscript proposes an AI-based automatic masking and inpainting process that focuses on the people in images. The approach involves segmenting the people in image with the help of a custom-trained DeepLabV3 + Network for masking them. Proposed approach further dilates the obtained mask followed by inpainting the masked region using the fast Fourier convolutions-based Large Mask Inpainting Model. The proposed solution shows acceptable results and potential for expansion to include numerous other classes of objects as well.