ASTAD: Asymmetric Style Transfer for Synthetic-to-Real Adaptation in Autonomous Driving

2026-06-28

AI Summary

This work addresses the performance degradation in autonomous driving caused by the domain gap between synthetic and real-world data, particularly when unlabeled real reference images hinder semantically consistent style transfer. To tackle this challenge, the paper introduces Asymmetric Style Transfer for Autonomous Driving (ASTAD), a novel task formulation, and proposes ASTModel, a training-free two-stage framework. Leveraging only labeled synthetic content images and unlabeled real style images, the method extracts coarse semantic priors and dynamically refines them during diffusion-based denoising to inject class-consistent styles, thereby avoiding semantic misalignment while eliminating the need for costly annotations. Experiments demonstrate that the approach significantly outperforms existing methods in downstream perception tasks and structural fidelity, achieving a 3.2× speedup in inference and substantially improving the practicality of synthetic-to-real domain adaptation for deployment.

Abstract

Synthetic data mitigates the data scarcity problem in autonomous driving perception. However, the synthetic-to-real gap leads to performance degradation, hindering real-world model generalization. Although current methods leverage diffusion models for photorealistic style transfer to bridge this gap, they critically ignore a practical asymmetry: while synthetic data possesses perfect pixel-level annotations, real-world style reference images generally lack corresponding labels. Consequently, existing methods relying on symmetric semantic guidance suffer from either prohibitive annotation costs or severe semantic misalignment. To address this dilemma, we formally propose a novel task: Asymmetric Style Transfer for Autonomous Driving (ASTAD), which requires semantically consistent transfer using only labeled synthetic content and unlabeled real-world references. We further introduce the ASTModel, a training-free two-stage framework designed to bridge this domain gap under asymmetric constraints. ASTModel first extracts a coarse semantic prior from the unlabeled target, followed by dynamic prior refinement and class-consistent style injection during the denoising process. Extensive experiments demonstrate that ASTModel significantly outperforms existing methods in downstream perception utility and structural fidelity, while offering a 3.2$\times$ inference speedup. This work aligns synthetic-to-real adaptation with practical constraints, holding the potential to accelerate the scalable deployment of robust autonomous driving systems. Code: https://github.com/Dingyi-Yao/ASTAD.

Problem

• synthetic-to-real adaptation
• asymmetric style transfer
• autonomous driving
• domain gap
• semantic misalignment

Innovation

• asymmetric style transfer
• synthetic-to-real adaptation
• diffusion models
• semantic prior refinement
• autonomous driving perception