Unpaired Image-to-Image Translation via a Self-Supervised Semantic Bridge

2026-02-18

AI Summary

Existing unpaired image translation methods exhibit limited generalization under adversarial training, while diffusion inversion approaches struggle to preserve structural fidelity. To address these challenges, this work proposes the Self-Supervised Semantic Bridge (SSB) framework, which leverages a self-supervised visual encoder to construct a shared latent space that is invariant to appearance variations yet retains geometric structure. This latent representation serves as a semantic prior for a diffusion bridge model, enabling spatially consistent and high-fidelity image translation without requiring cross-domain supervision. Experimental results demonstrate that SSB outperforms current state-of-the-art methods in medical image synthesis tasks, both in in-domain and out-of-domain settings, and further supports high-quality text-guided editing.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Unsupervised & Self-Supervised Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Adversarial diffusion and diffusion-inversion methods have advanced unpaired image-to-image translation, but each faces key limitations. Adversarial approaches require target-domain adversarial loss during training, which can limit generalization to unseen data, while diffusion-inversion methods often produce low-fidelity translations due to imperfect inversion into noise-latent representations. In this work, we propose the Self-Supervised Semantic Bridge (SSB), a versatile framework that integrates external semantic priors into diffusion bridge models to enable spatially faithful translation without cross-domain supervision. Our key idea is to leverage self-supervised visual encoders to learn representations that are invariant to appearance changes but capture geometric structure, forming a shared latent space that conditions the diffusion bridges. Extensive experiments show that SSB outperforms strong prior methods for challenging medical image synthesis in both in-domain and out-of-domain settings, and extends easily to high-quality text-guided editing.

Problem

• unpaired image-to-image translation
• adversarial diffusion
• diffusion-inversion
• low-fidelity translation
• generalization limitation

Innovation

• Self-Supervised Semantic Bridge
• unpaired image-to-image translation
• diffusion bridge
• semantic priors
• spatially faithful translation