arXiv 2601.037182026-01-08cs.CVeess.IVphysics.optics

通过领域适应利用未标记数据实现真实世界的镜头主动对齐

Towards Real-world Lens Active Alignment with Unlabeled Data via Domain Adaptation

organization= State Key Laboratory of Extreme Photonics；Instrumentation, College of Optical Science；Engineering , addressline= Zhejiang University , city= Hangzhou , postcode= 310027 , country= China；organization= National Research Center for Optical Instrumentation , addressline= Zhejiang University , city= Hangzhou , postcode= 310027 , country= China；organization= School of Artificial Intelligence；Robotics , addressline= Hunan University , city= Changsha , postcode= 410012 , country= China；organization= National Engineering Research Center of Robot Visual Perception；Control Technology , addressline= Hunan University , city= Changsha , postcode= 410082 , country= China；organization= Intelligent Optics \& Photonics Research Center, Jiaxing Research Institute , addressline= Zhejiang University , city= Jiaxing , postcode= 314031 , country= China；organization= Dongguan Yutong Optical Technology Co., Ltd. , city= Dongguan , postcode= 523863 , country= China；organization= Sunny Optical (Zhejiang) Research Institute Co., Ltd. , city= Hangzhou , postcode= 311215 , country= China

Wenyong Li, Qi Jiang, Weijian Hu, Kailun Yang, Zhanjun Zhang, Wenjun Tian, Kaiwei Wang, Jian Bai

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出DA3方法，通过领域适应利用未标记数据提升真实世界镜头主动对齐的精度和效率

摘要：

主动对齐（AA）是大规模自动化高精度光学系统组装中的关键技术。与劳动密集型的每台设备校准相比，基于光学模拟的数字双流水线在生成大规模标记数据方面具有显著优势。然而，复杂的成像条件导致模拟与真实世界图像之间存在领域差距，限制了模拟训练模型的泛化能力。为了解决这个问题，我们提出在模拟基线中加入少量在随机偏移位置捕获的未标记真实世界图像，从领域适应的角度来弥合这一差距。我们引入了领域自适应主动对齐（DA3），它利用自回归领域转换生成器和基于对抗的特征对齐策略，通过自监督学习来提炼真实世界领域的信息。这使模型能够提取领域不变的图像退化特征，以促进鲁棒的偏移预测。在两种镜头类型上的实验表明，DA3在纯模拟流水线上的精度提高了46%。值得注意的是，它接近通过在3个镜头样本上精确标记的真实世界数据获得的性能，同时将设备数据收集时间减少了98.7%。结果表明，领域适应有效地使模拟训练模型具备了鲁棒的真实世界性能，验证了数字双流水线作为显著提高大规模光学组装效率的实用解决方案。

英文摘要：

Active Alignment (AA) is a key technology for the large-scale automated assembly of high-precision optical systems. Compared with labor-intensive per-model on-device calibration, a digital-twin pipeline built on optical simulation offers a substantial advantage in generating large-scale labeled data. However, complex imaging conditions induce a domain gap between simulation and real-world images, limiting the generalization of simulation-trained models. To address this, we propose augmenting a simulation baseline with minimal unlabeled real-world images captured at random misalignment positions, mitigating the gap from a domain adaptation perspective. We introduce Domain Adaptive Active Alignment (DA3), which utilizes an autoregressive domain transformation generator and an adversarial-based feature alignment strategy to distill real-world domain information via self-supervised learning. This enables the extraction of domain-invariant image degradation features to facilitate robust misalignment prediction. Experiments on two lens types reveal that DA3 improves accuracy by 46% over a purely simulation pipeline. Notably, it approaches the performance achieved with precisely labeled real-world data collected on 3 lens samples, while reducing on-device data collection time by 98.7%. The results demonstrate that domain adaptation effectively endows simulation-trained models with robust real-world performance, validating the digital-twin pipeline as a practical solution to significantly enhance the efficiency of large-scale optical assembly.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。