The Power of Light: Improving Synthetic-to-Real Domain Adaptation through Physically-Based Indirect Illumination

The paper introduces SmartSDG, an automated pipeline utilizing NVIDIA Isaac Sim with Physically-Based Shading (PBS) for synthetic-to-real domain adaptation in object detection. It presents the ILLUM_INTRUCK dataset and demonstrates through 18 experiments with the YOLOv12 framework that complex indirect lighting and varied backgrounds enhance detection performance by preserving surface textures and reducing false positives. This research provides practical guidelines for optimizing virtual scene design, crucial for improving robustness in industrial automation applications.