This research introduces a novel approach to image correction using fuzzy-conditioned diffusion and diffusion-derived attention maps. The method leverages the power of image diffusion models to synthesize images and correct anomalies in facial images. The fuzzy conditioning allows for pixel-wise modification of different image components to varying degrees. The diffusion-derived attention maps estimate the degree of anomaly, leading to interpretable and autonomous facial image correction. The proposed method shows promise in the field of image synthesis and correction, providing a new way to control the degree of image conditioning and perform autonomous face correction.
Publication date: June 26, 2023
Project Page: https://github.com/majedelhelou/FC-Diffusion
Paper: https://arxiv.org/pdf/2306.14891.pdf
