Problem

When decimating point clouds, the decimate() function creates new smaller vectors and assigns them to the member vectors. However, std::vector assignment does not reduce capacity - it keeps the original memory allocation even when the new data is much smaller.

This meant that even with aggressive decimation, the application retained the full memory footprint of the original undecimated point clouds.

Solution

Added shrink_to_fit() calls after each vector assignment in PointCloud::decimate() to release unused memory back to the system.

Results

Tested on multi_view_tls_registration_step_2.exe with a 5km dataset using 0.2m decimation. Before the fix, the application used 80GB of RAM. After the fix, it uses only 10GB - an 88% reduction in memory usage. (Tested on Windows with msvc compiler)

Files changed

• core/src/point_cloud.cpp - Added shrink_to_fit() to 6 vectors in decimate()