Frequency-Aware Dual-Stream Learning for Balanced Realism and Fidelity in Electron Microscopy Imaging

2026-07-23

AI Summary

This work addresses the inherent trade-off between resolution and acquisition speed in electron microscopy imaging, where existing single-stream methods struggle to simultaneously achieve perceptual realism and quantitative fidelity, often resulting in oversmoothed details or artifacts. To overcome this limitation, we propose the first frequency-adaptive dual-stream architecture, which leverages the discrete wavelet transform to decouple an image into low-frequency structural components and high-frequency fine details. A conditional diffusion model generates globally realistic images from the low-frequency stream, while a Transformer network precisely restores high-frequency details. By introducing frequency-aware dual-stream processing into electron microscopy for the first time, our approach effectively disentangles the optimization of realism and fidelity. Evaluated on the EMDiffuse dataset, our method significantly outperforms state-of-the-art approaches in LPIPS and resolution metrics and demonstrates strong generalization across diverse biological specimens, enabling high-quality, rapid, and reliable imaging.

Abstract

Electron microscopy enables nanoscale cellular visualization but faces a trade-off between imaging resolution and acquisition speed. Existing learning-based methods rely on single-stream architectures that struggle to balance perceptual realism and quantitative fidelity, either over-smoothing details or generating unrealistic hallucinations. This work introduces a frequency-adaptive dual-stream architecture to resolve this conflict. Using discrete wavelet transform, we decompose images into low-frequency structures and high-frequency details, then employ a conditional diffusion model for realistic global synthesis and a transformer network for precise detail recovery. Experiments on the EMDiffuse dataset show the method achieves superior LPIPS and resolution ratio, substantially outperforming existing approaches. The method also shows strong generalization across diverse biological samples, supporting fast and reliable electron microscopy imaging for structural biology and nanotechnology applications. The source code and associated dataset are publicly available to facilitate further research.

Problem

• electron microscopy
• realism
• fidelity
• image reconstruction
• frequency decomposition

Innovation

• frequency-aware dual-stream
• conditional diffusion model
• transformer network
• discrete wavelet transform
• electron microscopy imaging