Data processing for bezier lane

[anhtr-nguyn]

Dear authors,

I do not understand why you transform the ground truth centerline into control points of beizer curve, then during the training phase, you transform these control points into 11 points of centerline again. What is the purpose of this process ? Why we dont just use ground truth directly (interpolate it into 11 points, for example)

Thank you

[wudongming97]

In the beginning, we use bezier curves because they are visibly smooth. However, from our latest experiments, regressing 11 points performs similarly to the bezier method.