iralabdisco / 3DSmoothNet

[CVPR2019] The Perfect Match: 3D Point Cloud Matching with Smoothed Densities

3DSmoothNet

1. Generate the Smoothed Density Valie (SVD) voxel grid and keypoints (saved as .pcd) using the voxelize_all.py. Modify the input and output direcotries at the beginning of hte file. The parameters for the SVD grid are stored in the voxel_parameters.yaml file. Do not modify the directory parameters in the .yaml file, they are update autmatically by the script.
2. Use the run_benchmark_global.py to generate the features from the SVD rapresentation.
3. Use the join_all.py script to merge keypoints and features in a single .npz file.

SLAM example

1. voxelize_tum_slam.py
2. run_benchmark_global.py (modify the )
3. join_feature_keypoints.py --cnn_dim 64 -f /benchmark/tum_slam_comparison/3DSmoothNet/features/pioneer_slam2_025/ -o /benchmark/tum_slam_comparison/3DSmoothNet/features/pioneer_slam2_025/