LBM is a scalable method for fast image-to-image translation using Bridge Matching in a latent space, achieving state-of-the-art results with a single inference step for tasks like object removal, normal and depth estimation, and object relighting.
