This is a nbv simulation system for comparative experiment, supporting our paper "A Global Max-flow-based Multi-resolution Next-best-view Method for Reconstruction of 3D Unknown Objects" (doi: 10.1109/LRA.2021.3132430). Armadillo_example_Ours directory contains an result example of our code with sampled viewsapce (the file 3d_models/Armadillo.txt).

For our nbv_simulation c++ code, these libraries need to be installed: opencv 4.4.0, PCL 1.9.1, Eigen 3.3.9, OctoMap 1.9.6. For nbv_net, please follow https://github.com/irvingvasquez/nbv-net. We tested our codes on Windows 10. For other system, please check the file read/write or multithreading functions in the codes.

Change "const static size_t maxSize = 100000;" to "const static size_t maxSize = 1000" in file OcTreeKey.h, so that the code will run faster.

1. Sample your 3d object model from *.obj or *.ply to *.pcd, and there is an example in directory 3d_models. For the sampling method, please follow https://github.com/PointCloudLibrary/pcl/blob/master/tools/mesh_sampling.cpp, and run with "./pcl_mesh_sampling.exe *.ply *.pcd -n_samples 100000 -leaf_size 0.5 -no_vis_result" or -leaf_size 0.0005, depending on the model.
2. Change the directory and model name in file DefaultConfiguration.yaml.
3. Put the file DefaultConfiguration.yaml in the correct path, and then run compiled program of main.cpp.
4. For the method_of_IG: Ours is 0, OA is 1, UV is 2, RSE is 3, APORA is 4, Kr is 5, NBVNET is 6. If you want to run with NBV-Net, please check nbv_net_path in file DefaultConfiguration.yaml, and run both compiled program of main.cpp and "python nbv_net/run_test.py {}" (replace {} by your model name) in pytorch environment at the same time.
5. There is a parameter "show", by default is 1, which means that the middle cloud will be shown in a pcl window, and close it to continue. If you don't want to show the middle cloud, change it to 0.

Please contact [email protected]