ResMatching: Noise-Resilient Computational Super-Resolution via Guided Conditional Flow Matching

TL;DR

The paper addresses the ill-posed nature of computational super-resolution in fluorescence microscopy by learning a data-specific structural prior. It introduces ResMatching, a guided conditional flow matching approach that enables posterior sampling and calibrated uncertainty, producing HR reconstructions conditioned on LR inputs. Empirically, it yields competitive data fidelity while delivering strong perceptual realism and reliable uncertainty estimates, even under challenging noise. This uncertainty-aware CSR framework enhances trust and decision-making in microscopic image analysis and potentially guides future refinement of priors for inverse imaging problems.

Abstract

Computational Super-Resolution (CSR) in fluorescence microscopy has, despite being an ill-posed problem, a long history. At its very core, CSR is about finding a prior that can be used to extrapolate frequencies in a micrograph that have never been imaged by the image-generating microscope. It stands to reason that, with the advent of better data-driven machine learning techniques, stronger prior can be learned and hence CSR can lead to better results. Here, we present ResMatching, a novel CSR method that uses guided conditional flow matching to learn such improved data-priors. We evaluate ResMatching on 4 diverse biological structures from the BioSR dataset and compare its results against 7 baselines. ResMatching consistently achieves competitive results, demonstrating in all cases the best trade-off between data fidelity and perceptual realism. We observe that CSR using ResMatching is particularly effective in cases where a strong prior is hard to learn, e.g. when the given low-resolution images contain a lot of noise. Additionally, we show that ResMatching can be used to sample from an implicitly learned posterior distribution and that this distribution is calibrated for all tested use-cases, enabling our method to deliver a pixel-wise data-uncertainty term that can guide future users to reject uncertain predictions.

Figures (4)

• Figure 1: ResMatching, using a conditional flow matching approach, leads to best-in-class computational super-resolution results even in severely noisy microscopy data.
• Figure 2: Quantitative results -- data fidelity and realism. Each row corresponds to results obtained on each data-subset we trained on (see Section \ref{['sec:experiments']}). We show PSNR vs. MicroMS-SSIM (left, note inverted axis for more consistent plots) and LPIPS vs. FID (center), capturing the trade-off between pixel-level fidelity and perceptual quality. ResMatching is highlighted with an additional red circle. Results tables (right) show the same data.
• Figure 3: Qualitative results. Representative predictions are shown for all data subsets (experiments). For each result (I-IV), we show: (a) the full low-resolution noisy input with a highlighted crop region (yellow box), (b) the cropped input region, (c) the high-resolution ground truth, (d–j) predictions of baseline methods (see Section \ref{['sec:experiments']}), (k) a single posterior sample generated by ResMatching, and (l) the MMSE estimate computed over 50 posterior samples.
• Figure 4: Model calibration. We show root mean variance (RMV) versus root mean square error (RMSE) as described in Section \ref{['sec:method']}. Each plot ((a)--(d)) corresponds to one experiment we conducted. The dashed line indicates $y=x$.