ApproximateProgressiveMorphologicalFilter: check for finite-ness #5756
Conversation
mvieth
added
module: segmentation
changelog: fix
| @@ -129,7 +130,9 @@ pcl::ApproximateProgressiveMorphologicalFilter<PointT>::extract (Indices& ground | |||
| for (int i = 0; i < static_cast<int>(input_->size ()); ++i) | |||
| { | |||
| // ...then test for lower points within the cell | |||
| PointT p = (*input_)[i]; | |||
| const PointT& p = (*input_)[i]; | |||
| if (!pcl::isFinite(p)) | |||
There was a problem hiding this comment.
Don't we need to test at line 235 as well - ground indices is the orignally indices input (so could be all points, if not thing is set), which results in a copy of the original cloud, iterated at the mentioned line?
There was a problem hiding this comment.
Should we only keep these checks for algorithms that require ordered point clouds and leave them out of algorithms, that works on unordered - to avoid the performance penalty of checking all points (and this case, twice?)
There was a problem hiding this comment.
However, there is no way to inform the user that its a ill-formed cloud, if we don't check them... so I guess it would bring these issues once in a while.
I guess the best option is to add a "is dense" version of all algorithms, where the user knows what is used as input.
Otherwise the matrices
AandZfmay be accessed with invalidrowandcolindices.Fixes #5728