Results of the evaluation of the leaf surface reconstruction in terms of robustness against noise and missing points. Each row (A–H) corresponds to the input point cloud for one leaf. Each pair of rows ((a) plus (b), (c) plus (d)) for a species corresponds to a split dataset with few and many missing points. Column 1 corresponds to the raw data for the input leaf point cloud, column 2 corresponds to the leaf point cloud transformed to the leaf coordinate system, columns 3–7 correspond to the results from two previous model-free methods (nonuniform rational B-spline (NURBS), Poisson, and methods from Zhu et al. [26
]), column 8 corresponds to the results from the flattening in the bending direction developed in the present study, and column 9 corresponds to results from the flattening in both the bending and rolling directions developed in the present study. Orange circles indicate where the missing points mostly exist in the leaf point cloud.