bug in calculating the filtered pointcloud after sample indexing

[rmessaou]

Hi Atenpas,

I am using gpd in my project and I had difficulties to work with the indexed pointcloud as input. It seems like the region that I am defining is not taken into consideration when the grasps are generated.

After a thorough look into the code, in the function where the filtering of the pointcloud using the indices is done, I noticed this:

gpd/src/gpd/util/cloud.cpp

Lines 206 to 222 in 6327f20

	void Cloud::filterWorkspace(const std::vector<double> &workspace) {
	// Filter indices into the point cloud.
	if (sample_indices_.size() > 0) {
	std::vector<int> indices_to_keep;
	for (int i = 0; i < sample_indices_.size(); i++) {
	const pcl::PointXYZRGBA &p = cloud_processed_->points[sample_indices_[i]];
	if (p.x > workspace[0] && p.x < workspace[1] && p.y > workspace[2] &&
	p.y < workspace[3] && p.z > workspace[4] && p.z < workspace[5]) {
	indices_to_keep.push_back(i);
	}
	}
	sample_indices_ = indices_to_keep;
	std::cout << sample_indices_.size()
	<< " sample indices left after workspace filtering \n";
	}

and specifically this line

gpd/src/gpd/util/cloud.cpp

Line 215 in 6327f20

indices_to_keep.push_back(i);

So it seems like the indices that are being kept are not "the sample indices" but actually their indices in the for loop list (so technically you are saving the index i instead of sample_indices_[i]).

Therefore shouldn't this line be instead like this

 indices_to_keep.push_back(sample_indices_[i]);

Can you confirm this ?
Thank you and best regards
Rayene

[anmakon]

Hi @rmessaou,
I'm a bit late to the game here, but your fix just saved my day and I can confirm it's working!
I think this bug could have been introduced somewhere along with the restructuring of the code in commit #83035e5, since I had earlier versions (with a weird version of ROS kinetic) working before without this issue.

Thank you!