Depth-Aware Unpaired Image-to-Image Translation for Autonomous Driving Test Scenario Generation Using a Dual-Branch GAN

Donghao Shi^1,2*Chenxin Zhao^1,2*Cunbin Zhao^1,2Zhou Fang^1,2Chonghao Yu^1,2Jian Li^1,2Minjie Feng^1,2

• ¹Advanced Manufacturing Metrology Research Center, Zhejiang Institute of Quality Sciences, Hangzhou, Jiangsu Province, China
• ²Key Laboratory of Acoustics and Vibration Applied Measuring Technology,State Administration for Market Regulation, Hangzhou, Jiangsu Province, China

Reliable visual perception is essential for autonomous driving test scenario generation, yet adverse weather and lighting variations pose significant challenges to simulation robustness and generalization. Traditional unpaired image-to-image translation methods primarily rely on RGBbased transformations, often resulting in geometric distortions and loss of structural consistency, which can negatively impact the realism and accuracy of generated test scenarios. To address these limitations, we propose a Depth-Aware Dual-Branch Generative Adversarial Network (DAB-GAN) that explicitly incorporates depth information to preserve spatial structures during scenario generation. The dual-branch generator processes both RGB and depth inputs, ensuring geometric fidelity, while a self-attention mechanism enhances spatial dependencies and local detail refinement. This enables the creation of realistic and structure-preserving test environments that are crucial for evaluating autonomous driving perception systems, especially under adverse weather conditions. Experimental results demonstrate that DAB-GAN outperforms existing unpaired image-to-image translation methods, achieving superior visual fidelity and maintaining depth-aware structural integrity. This approach provides a robust framework for generating diverse and challenging test scenarios, enhancing the development and validation of autonomous driving systems under various real-world conditions.删除了: This enables the creation of realistic and structure-37 preserving test environments, particularly for adverse weather 38 conditions, which are crucial for evaluating autonomous driving 39 perception systems. Despite the importance of real-world testing, collecting diverse adverse weather data remains costly, time-consuming, and logistically challenging. Moreover, real-world datasets often suffer from imbalance and limited coverage of extreme conditions, restricting their utility in comprehensive validation and robustness assessment (Agarwal et al., 2024;Lan et al., 2024;Li Y. et al., 2023). As a result, simulation-based testing has become a critical tool for autonomous driving development, allowing for the controlled generation of challenging environmental conditions to enhance model reliability and adaptability (Biagiola and Tonella, 2024;Huang et al., 2025;Sadid and Antoniou, 2024). A key requirement for effective simulation is the ability to generate photorealistic and geometrically consistent test scenarios that accurately reflect real-world conditions.