Bug: dense.cpp hardcoded values lead to indexing overflow in large point clouds.

[ashtanmistal]

In main.cpp:

for (int i = 0; i < pnumber; i++)
    {
        fscanf(pointreader, "%lf %lf %lf", &po[i].xyz[0], &po[i].xyz[1], &po[i].xyz[2]);
        int ppp = floor(((po[i].xyz[0] + 0.5) / cell)) * boxsize * boxsize + floor(((po[i].xyz[1] + 0.5) / cell)) * boxsize + floor(((po[i].xyz[2] + 0.5) / cell));
        search.push(ppp);
    }

If the pnumber is larger than the hardcoded maximum values for the KDtree, it will a) continue to try to push points to search, and no error is raised; b) no assertions are created to prevent inputting a point cloud that happens to be larger than this maximum, and c) nowhere in the code or paper is the limitation stated that the implementation only works for small (<5000 points) point clouds.

[xnowbzhao]

Thank you for your issue, it is prefered to divided large point cloud into blocks and apply upsampling respectively, I've modified the readme to point this out.