Structure of the CSV :

Frame: id of the frame in the form a-b where a is the frame number and is unique for every recording (sequence of frame or unique frame), and b is the frame sub count.

For a sequence of frames, b gives the number of the jth frame within the asequence for a single snapshot, b = 0

Landmark_ID : id of the node (see picture from media pipe) identifies part on the hand

Normalized_X Normalized_Y are the normalized positions of each node (-1,1 on the x-y plane)