发表机构
Center for Advanced Systems Understanding (CASUS); Helmholtz-Zentrum Dresden-Rossendorf e. V. (HZDR); Institute of Computer Science, University of Wrocław; School of Computation, Information and Technology, Technical University of Munich; Cluster of Excellence Physics of Life（高级系统理解中心; 德累斯顿-罗森多夫亥姆霍兹中心; 弗罗茨瓦夫大学计算机科学学院; 慕尼黑工业大学计算、信息与技术学院; 生命物理学卓越集群）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究针对科学成像中噪声数据难以获取干净参考的问题，提出NR-CFM模型，可从单张带噪图像学习生成干净样本，在多数场景优于NR-GAN，高噪声下与Ambient Diffusion相当，低信噪比科学数据上表现良好。
AI 中文摘要
生成模型学习训练数据的统计特性，因此高质量生成依赖于干净且具有代表性的数据集。在科学成像中，采集过程常产生带噪测量值，而收集干净参考数据可能成本高昂、不切实际甚至无法实现。直接在这些测量值上训练会得到重现损坏数据的模型。这一问题可通过直接从噪声数据中学习干净的总体分布来解决。条件流匹配（CFM）结合了简单的回归目标、稳定的训练过程、高效的采样能力以及强大的图像生成性能，使其成为该场景下的自然框架。我们提出噪声鲁棒条件流匹配（NR-CFM），这是一种从每张图像的一个损坏观测值中学习的无条件生成器。NR-CFM针对加性白高斯噪声提供了闭式干净端点校正，并针对具有更复杂协方差结构的一般高斯损坏学习数据驱动校正。在评估的所有损坏设置中，NR-CFM在大多数情况下性能优于NR-GAN，且在高噪声 regime 下与Ambient Diffusion具有竞争力。我们进一步在信噪比低至0.001的科学数据上评估NR-CFM，它能从严重损坏的测量值中生成合理的粒子图像。
英文摘要
Generative models learn the statistical properties of their training data, so high-quality generation depends on clean and representative datasets. In scientific imaging, acquisition often yields noisy measurements, while collecting clean references can be costly, impractical or even unattainable. Training directly on these measurements results in a model that reproduces the corrupted data. This can be circumvented by learning the clean population distribution directly from the noisy data. Conditional flow matching (CFM) combines a simple regression objective with stable training, efficient sampling, and strong image-generation performance, making it a natural framework for this setting. We introduce Noise-Robust Conditional Flow Matching (NR-CFM), an unconditional generator that learns from one corrupted observation per image. NR-CFM provides a closed-form clean endpoint correction for additive white Gaussian noise and learns a data-driven correction for general Gaussian corruptions with more complex covariance structure. Across the evaluated corruption settings, NR-CFM outperforms NR-GAN in most cases and remains competitive with Ambient Diffusion in the high-noise regime. We further evaluate NR-CFM on scientific data at signal-to-noise ratios as low as $0.001$, where it generates plausible particle images from severely corrupted measurements.
Comments10 pages, 3 Figures, and an Appendix
