Figure 2. A schematic of Residual Diffusion Bridge Models. RDBM utilizes a diffusion process guided by Doob’s h-transform towards an endpoint xT = µ free from stationary noise λϵ. Modulated by the residual component π = x0 − xT , the noise perturbation is selectively imposed on different regions with diverse degradation levels, thereby constructing probabilistic paths. Besides, it learns to reverse the process by matching the residual bridge score functions, facilitating an adaptive inversion from xT ∼ pprior(x) to x0 ∼ pdata(x).