ive been working on a sign language recognition. i extracted landmarks with mediapipe, saved it as .parquets then padded the data to create uniform length. each row of landmark has 21 node with x,y,z coordinate as shown. the padding done according to max_frame_size i found among all videos (311).