Fig. 2.

Video analysis using deep learning algorithms for tapping protocol (associated Supplementary Movie 1). a) Snapshots from the initial videos, with the different skeletons extracted using the two deep learning algorithms (blue DeepLabCut 2D, red HandGraphCNN 2D, green HandGraphCNN 3D). b1, left) Evolution of the distance between the thumb and pinky fingers tips with 9 sweeps (delimited by the 10 plain circles on each trajectory). b1, right) First three sweeps (dotted lines) compared to the last three ones (plain lines). b2) same as (b1) for the speed. c) Frequency F, fatigue L, relative variation of the period ΔT_sweep, amplitude ΔA_sweep, speed ΔS_sweep accross the protocol, more complex variation coefficients, ΔT_fit and ΔA_fit based on the fit of a periodic function.