Image Relighting

Image relighting aims to realistically change the illumination of a scene depicted in a single image, simulating different lighting conditions. Recent research focuses on deep learning approaches, employing diffusion models and networks designed to incorporate physical lighting properties, often leveraging synthetic datasets for training and evaluation. These advancements enable more accurate and versatile relighting, with applications ranging from augmented reality object insertion to data augmentation for autonomous driving, improving the realism and robustness of computer vision systems. The development of large-scale, high-quality datasets is also a key area of focus, facilitating the training of more powerful and generalizable relighting models.