DBMSolver: A Training-free Diffusion Bridge Sampler for High-Quality Image-to-Image Translation

TL;DR

DBMSolver is a training-free diffusion sampler that significantly improves efficiency and quality in image-to-image translation tasks by exploiting the structure of diffusion models, enabling practical real-world applications.

Contribution

It introduces a novel, training-free diffusion sampler using exponential integrators that reduces sampling evaluations and enhances image quality in diffusion-based I2I translation.

Findings

Reduces NFEs by up to 5x while improving FID scores.

Achieves state-of-the-art efficiency-quality tradeoffs across multiple tasks.

Enables real-world applications with high-resolution image translation.

Abstract

Diffusion-based image-to-image (I2I) translation excels in high-fidelity generation but suffers from slow sampling in state-of-the-art Diffusion Bridge Models (DBMs), often requiring dozens of function evaluations (NFEs). We introduce DBMSolver, a training-free sampler that exploits the semi-linear structure of DBM's underlying SDE and ODE via exponential integrators, yielding highly-efficient 1st- and 2nd-order solutions. This reduces NFEs by up to 5x while boosting quality (e.g., FID drops 53% on DIODE at 20 NFEs vs. 2nd-order baseline). Experiments on inpainting, stylization, and semantics-to-image tasks across resolutions up to 256x256 show DBMSolver sets new SOTA efficiency-quality tradeoffs, enabling real-world applicability. Our code is publicly available at https://github.com/snumprlab/dbmsolver.