arXiv subjects

Dan Fu

Publications and source records attributed to Dan Fu.

4 recordsLinked to original sources

arXiv2026

MicroDiffuse3D: A Foundation Model for 3D Microscopy Imaging Restoration

Chemical imaging enables label-free visualization of cells, tissues and living systems while providing direct biochemical information that is difficult to obtain with conventional fluorescence microscopy. Despite its promise in applications ranging from intraoperative diagnosis to drug-response analysis, its broader use remains limited by slow data acquisition, particularly for three-dimensional imaging. Here we present MicroDiffuse3D, a pretrained foundation model for 3D microscopy image restoration that recovers high-quality volumetric structure from degraded low-resolution measurements acquired at substantially higher throughput. We evaluated MicroDiffuse3D across three challenging restoration settings, including 3D super-resolution under 16-fold volumetric sparsity, joint degradation in resolution and noise, and 3D denoising in the low signal-to-noise ratio (SNR) regime, where the model delivered clear gains over strong baselines. Under the sparse 3D super-resolution setting, MicroDiffuse3D produced clearer continuity across depth with fewer artifacts and improved segmentation quality by 10.58% and line-profile concordance by 15.59%. Together, our results establish pretrained 3D restoration as a broadly applicable strategy for overcoming the throughput and SNR limitations in volumetric chemical imaging, enabling high-resolution analysis at scales and speeds that were previously difficult to achieve.

cs.CV↗

arXiv2025

Confronting Contemporary Seasonality Changes in East Asian Tropical Cyclone Landfalls with a Multi-Century Historical Baseline

Paleoclimate records provide a critical long-term perspective on natural climate variability, essential for understanding contemporary climate change. However, existing paleoclimate proxies lack the spatial-temporal coverage for studying changes in high-impact weather extremes like tropical cyclones (TCs). Here we introduce a multi-source framework that confronts the contemporary changes in TC landfalls in East Asia with a multi-century baseline (1368-1911) reconstructed from historical documents. Leveraging pre-industrial and contemporary data, the analysis reveals that a relatively small shift toward earlier landfalls in the contemporary era (1946-2020). However, this shift falls well within the range of natural fluctuations documented historically (1651-1900). This low signal-to-noise ratio indicates the forced anthropogenic signal of TC landfall timing remains challenging to detect. Besides providing a template for assessing seasonality changes in extremes, our work shows consistent natural controls of TC timing in contemporary and pre-industrial eras, lending credibility to pre-industrial observational datasets and climate simulations.

physics.ao-ph↗

arXiv2023

ChiMera: Learning with noisy labels by contrasting mixed-up augmentations

Learning with noisy labels has been studied to address incorrect label annotations in real-world applications. In this paper, we present ChiMera, a two-stage learning-from-noisy-labels framework based on semi-supervised learning, developed based on a novel contrastive learning technique MixCLR. The key idea of MixCLR is to learn and refine the representations of mixed augmentations from two different images to better resist label noise. ChiMera jointly learns the representations of the original data distribution and mixed-up data distribution via MixCLR, introducing many additional augmented samples to fill in the gap between different classes. This results in a more smoothed representation space learned by contrastive learning with better alignment and a more robust decision boundary. By exploiting MixCLR, ChiMera also improves the label diffusion process in the semi-supervised noise recovery stage and further boosts its ability to diffuse correct label information. We evaluated ChiMera on seven real-world datasets and obtained state-of-the-art performance on both symmetric noise and asymmetric noise. Our method opens up new avenues for using contrastive learning on learning with noisy labels and we envision MixCLR to be broadly applicable to other applications.

cs.CE↗

arXiv2023

Impact of Climate Simulation Resolutions on Future Energy System Reliability Assessment: A Texas Case Study

The reliability of energy systems is strongly influenced by the prevailing climate conditions. With the increasing prevalence of renewable energy sources, the interdependence between energy and climate systems has become even stronger. This study examines the impact of different spatial resolutions in climate modeling on energy grid reliability assessment, with the Texas interconnection between 2033 and 2043 serving as a pilot case study. Our preliminary findings indicate that while low-resolution climate simulations can provide a rough estimate of system reliability, high-resolution simulations can provide more informative assessment of low-adequacy extreme events. Furthermore, both high and low-resolution assessments suggest the need to prepare for severe blackout events in winter due to extremely low temperatures.

physics.ao-ph↗