Training Dynamic Plane Convolutional Occupancy Network With Triplet Loss python train_point_plane.py triplet_fc_plane.yaml model_name You should specify margin_triplet in training part. Choose triplet_fc_plane_net as class model in encoder, cbatchnorm in decoder