Capacity of the annotation module. Assessment of the annotation module accuracy by comparing state estimations (either probabilistic or categorical) with ground truth from simulated tracks. If not stated otherwise, $d0*$ = 0, $d1*$ = 5, and k_u = k_b = 0.1 Δt⁻¹. (a) Fraction of time points actually in diffusive state depending on the probability to be diffusive estimated by ExTrack. More specifically, time points are binned according to their probability to be diffusive (x axis) and for each bin we computed the fraction actually in diffusive state (y axis). Binning of 0.01. (b–d) Categorical state predictions are obtained by picking the state with highest probability for each time point. The fraction of mislabeled time points can thus be computed by comparison to the known true states of the simulated tracks. (b) Heatmap of the fractions of mislabeled time points depending on $d1*$ and k (10,000 tracks of 10 time points). (c) Fraction of mislabeled time points depending on (temporal) distance to transition time points. (d) Fractions of mislabeled time points using correct parameters except for one of them specified in the legend. x-axis: Relative error of the varied parameter compared to the true value underlying the simulated tracks. For this particular simulation, we used 10,000 tracks of 20 time points. Window length of 10. (e) Fraction of mislabeled time points depending on $d1*$ for ExTrack and vbSPT. The grey dotted curve (annotated as Error dif.) is the relative difference of the fraction of mislabeled time points between vbSPT and ExTrack (error_vbSPT–error_ExTrack)/error_ExTrack (10,000 tracks of 10 time points).