(A) Schematic shows how the pose and orientation of a freely moving animal change with time. Black dot indicates head. (B) Pipeline to evaluate the fDNC model at tracking neurons within an individual across time. The fDNC model takes in positional features of a template neuron configuration from one time t1 of a freely moving worm, and predicts the correspondence at another time t2, called the test. Recording is of a moving animal undergoing calcium imaging from Nguyen et al., 2017. Ground truth neuron correspondence are provided by manual human annotation. The same time point is used as the template for all 1513 template-test pairs. (C) Performance of fDNC and alternative models at tracking neurons within an individual are displayed in order of mean performance. CPD refers to Coherent Point Drift. NeRVE(1) refers to the restricted NeRVE model that has access to only the same template as CPD and fDNC. NeRVE(100) refers to the full NeRVE model which uses 100 templates from the same individual to make a single prediction. A Wilcoxon signed rank significance test of fDNC’s performance compared to CPD, NeRVE(1) and NeRVE(100) yields and , respectively. Boxplots show median and interquartile range. (D) fDNC tracking performance by neuron. Cumulative fraction of neurons is shown as a function of the acceptable error rate. (E) Detailed comparison of fDNC tracking to human annotation of a moving GCaMP recording from Nguyen et al., 2017. Color at each time point indicates the neuron label manually annotated by a human. White gaps indicate that the neuron is missing at that time point. In the case of perfect agreement between human and fDNC, each row will have only a single color or white.