Human Relighting

Human relighting aims to realistically change the illumination of a scene or object in an image or video, achieving effects like shadow removal or altering the lighting environment. Current research heavily utilizes deep learning, particularly neural radiance fields (NeRFs) and deep unfolding networks, often incorporating physically-based rendering models and intrinsic image decomposition (separating albedo, shape, and lighting) to improve realism and efficiency. This field is significant for applications in augmented reality, virtual production, and image editing, offering advancements in both image manipulation and our understanding of light-matter interaction.