fix octree bounds reducing

[pkuhto]

for some resolution/bounds ratio octree bounds can be unexpectedly reduced

[jspricke]

Interesting, can you add a unit test with an example?

@jkammerl can you have a look?

[pkuhto]

const double someResolution(10);
const int someDepth(4);
const double desiredMax = ((1<<someDepth) + 0.5)*someResolution;
pcl::octree::OctreePointCloud<PointT> tree(someResolution);
tree.defineBoundingBox(0, 0, 0, desiredMax, desiredMax, desiredMax);
double min_x, min_y, min_z, max_x, max_y, max_z;
tree.getBoundingBox(min_x, min_y, min_z, max_x, max_y, max_z);
assert(max_x >= desiredMax && max_y >= desiredMax && max_z >= desiredMax);

of course bounds can be corrected in adoptBoundingBoxToPoint with treedepth=srcTreeDepth+2 (worst scenario), but I didn't use bounds correction for perfomance reasons and got points in invalid leaves.

[jspricke]

Can you put this into a unit test and add it to the pull request? That would be great :).

[SergioRAgostinho]

Nice catch. 👍

1. Please follow our coding style guide, more specifically notice how the spacing is used before parenthesis.
2. @jspricke I'm not sure how those ref_values in the GFPFH test were originally calculated, but I have the impression they relied on the bugged implementation of the octree, especially considering it's using the bunny pointcloud and not something much more synthetic that could be manually computed. Can we consider to update them?

[jspricke]

@SergioRAgostinho I guess they where computed and then later manually verified. If we don't change GFPFH in between, putting in the newly computed values should be fine. @pkuhto can you add that to the pull request as well? Thanks!

[SergioRAgostinho]

LGTM 👍 Squash all the commits into a single one and this should be good to go.

[taketwo]

Please use ASSERT_DOUBLE_EQ to make assertions about floating-point numbers.

[taketwo]

Squash all the commits

@SergioRAgostinho note that Github can do this. You will get an option to squash commits before merging when you press "Merge pull request" button.

[SergioRAgostinho]

@taketwo Should we squash-merge this and then just submit a minor corrections to those asserts?

[taketwo]

Sure. Please merge and I'll make a correction PR.