Here is the SHM code.
The code is tested with Pytorch 1.6.0 with CUDA 10.2.89. Prerequisites include scipy, h5py, tqdm, etc. Your can install them by yourself.
The ModelNet40 dataset, 3DMatch and KITTI datasets can be download from:
https://github.com/WangYueFt/dcp;https://github.com/yewzijian/RPMNet;https://github.com/chrischoy/DeepGlobalRegistration
Start training with the command:
python main.py
Start testing with the command:
python main.py --eval True --mdoel_path YOUR_CHECKPOINT_DIRECTORY
Start finetuen with the command:
python main.py --finetune True --tune_path YOUR_CHECKPOINT_DIRECTORY
The code is insipred by DCP, PRNet, RPMNet, DGR, etc.