BBDM: Image-to-image Translation with Brownian Bridge Diffusion Models

TL;DR

This work tackles cross-domain image-to-image translation by reframing translation as a bidirectional Brownian bridge diffusion process operating in a pretrained latent space. The Brownian Bridge Diffusion Model (BBDM) maps domain A latent codes to domain B latent codes, starting the reverse process from the target endpoint rather than using conditional inputs during inference. The method leverages a latent diffusion backbone (VQGAN/LDM) and an ELBO-based training objective with noise prediction, plus DDIM-style accelerated sampling. Empirical results across semantic synthesis, edges-to-photo, and style-transfer tasks show competitive FID and LPIPS scores with enhanced diversity, and ablations validate the robustness to latent spaces and sampling settings. The approach offers a conditioning-free alternative to conditional diffusion for robust, multi-domain image translation in latent space, with potential extensions to multi-modal tasks.

Abstract

Image-to-image translation is an important and challenging problem in computer vision and image processing. Diffusion models (DM) have shown great potentials for high-quality image synthesis, and have gained competitive performance on the task of image-to-image translation. However, most of the existing diffusion models treat image-to-image translation as conditional generation processes, and suffer heavily from the gap between distinct domains. In this paper, a novel image-to-image translation method based on the Brownian Bridge Diffusion Model (BBDM) is proposed, which models image-to-image translation as a stochastic Brownian bridge process, and learns the translation between two domains directly through the bidirectional diffusion process rather than a conditional generation process. To the best of our knowledge, it is the first work that proposes Brownian Bridge diffusion process for image-to-image translation. Experimental results on various benchmarks demonstrate that the proposed BBDM model achieves competitive performance through both visual inspection and measurable metrics.

Figures (13)

• Figure 1: Comparison of directed graphical models of BBDM (Brownian Bridge Diffusion Model) and DDPM (Denoising Diffusion Probabilistic Model).
• Figure 2: Architecture of BBDM.
• Figure 3: Qualitative comparison on CelebAMask-HQ dataset.
• Figure 4: Qualitative comparison on different image-to-image translation tasks.
• Figure 5: Diverse samples of BBDM on different image-to-image translation tasks.