search

humanpose1 / MS-SVConv

Compute descriptors for 3D point cloud registration using a multi scale sparse voxel architecture
MIT License
63 stars 7 forks source link

Can i build MS-SVConv model? #8

Closed jihoon9208 closed 2 years ago

jihoon9208 commented 3 years ago

Dear humanpose1. Thank you so much for revealing the tremendous code. I am working on feature extraction, and i am curious about feature extraction that MS-SVConv is using Can I make MS-SVConv model using various epochs?

Once again, thank you for your big decision.

humanpose1 commented 3 years ago

Hi @jjPenguin , Thank you for your warm message. I am not sure to fully understand your request but if I understand well, you are asking whether you can train multiple MS-SVConv models with various epochs. With torch points3d you can. For example, if you want to train on modelnet.

python train.py task=registration models=registration/ms_svconv_base model_name=MS_SVCONV_B2cm_X2_3head data=registration/modelnet_sparse_ss training=sparse_fragment_reg tracker_options.make_submission=True training.epochs=2

It will save a pt file on a specific folder. then you can resume training with a different number of epoch (you have tospecify also the checkpoint_dir)

python train.py task=registration models=registration/ms_svconv_base model_name=MS_SVCONV_B2cm_X2_3head data=registration/modelnet_sparse_ss training=sparse_fragment_reg tracker_options.make_submission=True training.epochs=4 training.checkpoint_dir=/path/to/directory/that/contain/pt/file

Does it help you ?