Answers are grounded in this paper and cite their sources.
Demonstrates improved 3D dislocation reconstruction in rapid-scan bright-field STEM, indicating enhanced image quality.
Rapid-scan bright-field STEM enables sub-second observation of dislocation motion, but low-dose acquisition introduces severe noise and tilt-dependent intensity variations that undermine cross-tilt geometric consistency. We propose a supervised denoising framework without slow-scan ground truth. First, we construct high-fidelity proxy references by combining BM3D pre-denoising with quality-weighted, drift-corrected frame integration, producing 552 proxy stacks from 12,168 rapid-scan frames. Second, we design a multi-scale residual network tailored for robustness across wide tilt ranges. Systematic comparisons show that models optimized on matched static data can degrade under tilt-series domain shift, while the proposed multi-scale design preserves cross-tilt consistency and supports feature-based stereographic reconstruction of dislocation-line geometry rather than volumetric tomography.
No takes yet. Share an insight, caveat, or question.
Bo et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: