In the paper part 2.4, it is said that ''regress the 3D coordinates via a series of fully connected layers on the feature of each point". However, in the function get_gen_model(), model_generator2_2new6.py, line 70 to 81, there are only conv2d layers?