avoid bspline

Hi, I'm finetuning your pretrained model on architectural point cloud. Because my data have almost no b-spline surface, I don't want the model to predict b-spline. How can I do that?

I find your definition of model has 'num_primitives' argument, will this variable be useful? Or would it be better to refit the sub-pointcloud with other primitives that is predicted as a B-spline patch?