We train PlainGCN-28 and ResGCN-28 models on one Tesla V100. For DenseGCN, we use 4 Tesla V100s.
For training ResGCN-28, run:
python main.py --phase train --n_blocks 28 --block res --data_dir /path/to/modelnet40
Just need to set --data_dir into your data folder, dataset will be downloaded automatically.
Models can be tested on one 1080Ti.
Our pretrained models are available Google Drive.
Use the parameter --pretrained_model to set a specific pretrained model to load. For example,
python main.py --phase test --n_blocks 28 --block res --pretrained_model /path/to/pretrained_model --data_dir /path/to/modelnet40