Fig. 4

DHPSFU performance under challenging conditions. (A) The algorithms’ sensitivity and precision (defined as in Fig. 2) at different average signal-to-noise (S/N) ratios and localisation densities. (B) Examples of frames from a low-S/N dataset (left; S/N = 5.8, 5 localisations per frame) and a high-density dataset (right; S/N = 23, 15 localisations per frame), illustrating the splitting or merging of localisations by SMAP (blue arrows) and failure to detect near-horizontal PSFs by EasyDHPSF (orange arrows). Scale bars, 1 μm. (C) The 3D localisation coordinate precision reached by each algorithm in different conditions, defined as the median distance between each detected localisation and its true position. (D) The runtime of each algorithm for two datasets with intermediate S/N, 5 or 15 molecules per image frame on average, and 5,000 frames. The darker and lighter colours show the portion of time necessary for the first (GDSC PeakFit for DHPSFU, template matching for EasyDHPSF) or the second (DHPSFU algorithm itself, double-Gaussian fitting for EasyDHPSF) step of the method, respectively. The times taken for calibration are not included.