Support Website Contact Support Blog

Point Cloud Classification

I hope I’m just missing something, but why would you remove point cloud classification in its entirety for version 3.1.18? We all know that no DTM surface is going to be perfect. How are you going to distinguish a road salt building (should be excluded from a DTM) and a stockpile (exact same shape, but should be included). Now if I want to manually edit the point cloud, I have to start from sqaure one and manually classify the entire scene? The classification obviously still goes on behind the scenes, PLEASE bring this functionality back!

I have run a project over the weekend and came in this morning to find there is no classification done. I would have thought that if the classification process is automatic it would be putting features into terrain and object layers within the point cloud?

If the classification process is automatic I don’t think we need the point cloud setting though, providing it can still be edited manually afterwards which by the looks of it you can.

The problem is that no DTM they generate fully automatically will be 100% correct. They might be ‘good’ but they won’t be perfect. There’s always tinkering that’s needed to generate an acceptable product for the client. This is close to a deal-breaker for me to shift to a different platform that is implementing classification more robustly. This new version feels like it’s getting a little too hand-holdy with less power for the end-user and more blind trust in black box algorithms.

Case in point: Just did this site. While it did an “ok” job…there’s at least 30 major aberrations like this (see attached massive sink hole generated in the middle of a rail yard), to go along with countless smaller issues that would be grossly apparent on a 50cm contour map. Sure, if I wanted a quick and dirty 2 or 5m contour map of an area, great…but that’s literally the exact opposite reason of why we fly UAVs to capture topography.

If I still had a Terrain / Object classification, I could simply fix the couple of points that are obviously creating these effects and move on. But now, I have to start from scratch and classify the whole scene, deleting objects, to create a DTM to a reasonable standard.

I’m not asking for perfect DTM generation…that’s not possible. I’m asking for the black box to be opened slightly so that we can at least massage the results to avoid having to start from square one.

Dear All,

It is true that the point cloud classification is temporarily removed from Pix4Dmapper. At the same time the automatic DTM generation is out of beta and gives good results. 

At the moment, we are developing a new, more robust point cloud classification that will classify automatically the points into groups in a much more efficient way than the algorithm of version 3.0 that was also in beta. These groups and any manual intervention will  also be taken into account for the DTM generation in step 3. This new algorithm will, most probably, be launched early 2017. 

For the moment, it is true that there is not a lot of space for manual intervention in the DTM generation. What you can do is edit the point cloud deleting the areas that are introducing noise to the DTM. The points assigned to the “Deleted” point group will not be taken into account for the DTM generation. 

We are also very interested in investigating why the DTM generation is not giving the expected results. We would appreciate it if you could share a dataset that does not give good results with us to further investigate and improve the DTM algorithm. If you wish, you can upload a project to our Cloud and let us know when the project is uploaded.

Best regards,

I, too, came in this morning expecting to classify my points only to find that feature had been removed! Extremely bummed! Something that works okay but not great is better than not having it at all. Especially in this case. In the past it worked well enough for me so all I had to do was a little clean up. Now I have to do it all.

Hopefully “early 2017” includes the month of February…

Please bring the feature back!  Even if it’s beta.  I’ll sign a waiver.  It worked well enough to eliminate most of the manual removal from a project.  I tried the DTM generation feature on a dataset last night and came back in to find that nothing was classified.  Buildings and vegetation were all still in place with no point group assigned to the classified points that should have been generated through the DTM process.  If you would like to see the dataset, let me know where to upload it, and what I need to upload - just the project file, or the whole project - and I’ll make it available.  

In the interim, is there any way to download the previous version that still has this tool in place, and if there is, will projects updated to the newest version of Mapper still be functional in that environment?

Chris - I’ve been saving previous versions of the install files for Pix4D since they don’t make them available. You can download them from my DropBox account, which is at the following link:

 

https://www.dropbox.com/sh/vwb95ddce46hf7k/AAAPzOPEcSJPO_eSVER3QMV5a?dl=0

 

Mark,

I love you.  

For what it is worth, I have been archiving releases for some time as well if for some reason even older releases are needed…

Also…

The terrain classification feature in this software has been developing but slow. I feel the pain too!

 Yeah.  It’s super slow.  The most frustrating thing, for me is that I can’t choose what to keep and what not to.  I have small piles of material in one project that I need to have represented, but that are removed with the vegetation.  I don’t understand why they couldn’t just leave the tool that was there in place.  Saved me a lot of time.

Yup!

We have requested to be able to put different classified data into groups so we can choose those groups or sub groups to be rendered out and analyzed. For now, we move groups in and out of the delete group in order achieve desired effects.

Always having to do work-around for bugs and unimplemented work-flows.

Hello all,

We understand that the point cloud classification is an important feature and therefore a new improved tool will be included in the software soon. Our developers are working on the integration of the tool in Pix4D Desktop. However, we cannot guarantee about the integration time as the feature is still under development. 

In the meantime, you can still edit the point cloud assigning points in the Deleted point group in order to improve the DTM generation. Please feel free to share with us any special projects that the DTM algorithm was not properly applied by submitting a Support request.

Regards,