unsatisfactory result in SPAwN for NYUCAD dataset

I train the SPAwN network following the readme, before data preparation, 2D network training and 3D preprocessing. My command for 3D Preprocessing is python spawn_preproc.py NYUCAD R213_BiModalRDFNetLW_NYU-RGB-NORMALS-FINE_EPOCH_128 My command for train is python spawn_train.py NYUCAD --alpha=0 --data_aug=yes

But my best val MIou% is 45.3, far away from 62.2 mentioned in your paper. I am wondering about the reason.