runyiyang/graphdownsampling

A Python API for large-scale graph-based method point cloud downsampling.

Compilation Instructions for Shared Object File

1. Compilation for .so File

Navigate to the graphScore directory and use the following command to compile the shared object file:

cd graphScore
g++ -shared -o filtergraphv2.so pyGraphFilterV2.cpp graphFilter.cpp pccProcessing.cpp -I./ -I/usr/local/include/eigen3/ -O3 -fopenmp -fPIC -Wall

2. Running the File

Execute the script using Python with a specified PLY file:

python test_ctypes.py ply/point_cloud.ply

Runtime for 5 Million Points:

Time taken to read ply:  0.3337249755859375 seconds
Processing 5339063 points with sample rate of 0.1
x1: 2.07133 y1: 2.02422 z1: -1.16529
x2: 1.48653 y2: 1.41477 z2: 1.65127
x3: 1.43754 y3: 1.37962 z3: 1.16425
Point Cloud size in C++: 5339063
Time: numpyArrayToDoubleArray and loadBlock cost 318 ms
Time: findNNdistances cost 12012 ms
Time: buildGraphEigen cost 41614 ms
Time: graphFilter cost 1388 ms
Time: computeScore cost 55037 ms
graphfilter.computeScore cost:  55.42829728126526 seconds
Time taken to sample:  56.352569818496704 seconds

3. Adding to the Conda Environment

Find Python Include Path:

Execute the following command to find the Python include path:

python -c "from distutils.sysconfig import get_python_inc; print(get_python_inc())"

Find NumPy Include Path:

To find the include path for NumPy, use:

python -c "import numpy; print(numpy.get_include())"

Compiling Version 1:

Compile the V1 version with the command below, ensuring the include paths are correctly specified for your Conda environment:

g++ -shared -o filtergraphv1.so pyGraphFilter.cpp graphFilter.cpp pccProcessing.cpp -I./ -I/usr/local/include/eigen3/ -I/path/to/conda/env/include/python3.7m/ -I/path/to/conda/env/lib/python3.7/site-packages/numpy/core/include/ -O3 -fopenmp -fPIC -Wall

The V1 version works without issues for cubic.ply but shows incorrect point cloud information for point_cloud_500w.ply, suggesting a problem in the data processing.

Correct Output:

x1: 2.07133 y1: 2.02422 z1: -1.16529
x2: 1.48653 y2: 1.41477 z2: 1.65127
x3: 1.43754 y3: 1.37962 z3: 1.16425

Current Output for V1:

x1: 0.0435803 y1: 0.0458223 z1: 0.0904121
x2: 0.0496849 y2: 0.662647 z2: 0.63962
x3: 0.532232 y3: 0.37154 z3: 0.397041
x4: 0.163577 y4: 0.193415 z4: 0.293161

3. Refer to Fast Resampling of 3D Point Clouds via Graphs

A python implement to https://arxiv.org/abs/1702.06397.