Fig. 3: Numerical comparison of sampling methods.
From: Demonstration of an AI-driven workflow for autonomous high-resolution scanning microscopy
A shows the ground truth with the color scale representing the normalized intensity, B–D show respectively the raster grid (RG), low-discrepancy random (LDR), and FAST reconstructions at 10% scan coverage, and G–I show the actual scan points that produce these reconstructions. E, F show the evolution of the normalized root mean square error (NRMSE), for which lower is better, and the Structural Similarity metric (SSIM), for which higher is better, as a function of the scan coverage. The FAST reconstruction stabilizes at 27% coverage, while the other techniques take significantly longer to reach the same quality. Source data are provided as a Source Data file.
