Fig. 4: Quantification of segmentation accuracy.

a Colour plot showing the segmentation accuracy as a function of \(\tilde{D}\) and \(\mathop{\tau }\limits^{ \sim }\), obtained after averaging over N_t = 100 trajectories of length T = 2000. b The overlap between the displacement distribution of the two states θ₁₂ and c the segmentation accuracy, both plotted as a function of the filter width f for the same values of \(\widetilde{D}\) and \(\widetilde{\tau }\) as in the other panels. d The accuracy as a function of the trajectory length T for various \(\widetilde{D}\) and \(\widetilde{\tau }\), keeping fixed the total number of sampled displacements N_t × T = 20000. e The accuracy as a function of the localisation error \(\widetilde{\sigma }\) for the same values of \(\widetilde{D}\) and \(\widetilde{\tau }\) as in panel (b). f The accuracy as a function of the intensity of motion blur, as mimicked by varying n in Eq. (3). In this case, the “ground truth” when determining the segmentation accuracy for the blurred trajectory was determined through a majority vote of all n points involved in the blurring process. Where applicable, shadings represent one standard deviation of the accuracy over N_t = 100 trajectories.