CloudCompare forum

And it is WAY faster!!
Between 3.5x and 121x (!!) faster from the original version (using the same parameters than for the previous comparison)
Really great!

Hello, our manuscript about CSF has been officially accepted, and is is ready to be publised. All the details can be found in this paper. We will add the reference information in the plugin as soon as possible. In addtion, we have ackowledged Daniel for his help of improving software performance. we will also continue to update this plugin and make this plugin better with the help from all of you.

hi, Thilo Forster. We replied your email, but it was refused by your mail server. So we put the news here, hope you can see it.

The paper is published and is uploaded to researchgate (https://www.researchgate.net/profile/Wuming_Zhang2). The title is "An Easy-to-Use Airborne LiDAR Data Filtering Method Based on Cloth Simulation". Please check it. If it is useful, please cite it in your work.
We will also update CSF plugin in several days to improve its efficiency.

Cheers

Hello, daniel. These days, we are working on improve this plugin. We tried nanoflann. It works well in stand alone application, as fast as the plugin using CC's octree. But when we integrated it into the newest plugin code, it ran very slowly. Certainly, we will try to debug it ASAP.
BTW, we tested the version using CC's octree, it performed quite well. Why it is abandoned? We didn't understand your post clearly. Could you please explain it in another way? Maybe it is easy to understand with an example. Thank you.

----------------------------------------------------------------------------------------------------------------------------------------------------
In fact I overlooked the possibility that the cloud could be quite far from the initial (cloth) grid. This was resulting in very far neighbor search and my code was not able to handle this properly. Waiting for a better approach, I reactivated the old code (based on CGAL) for the 'rasterize' part.

The issue only occurs when the cloud is very sparse (for instance the points are only grouped on two borders of the bounding-box (in a 'L' shape).

As the algorithm looks for the nearest neighbor of each grid points, this nearest point may be very far from the grid point. When this distance is very big, this is a worst case scenario for the DgmOctree structure. And therefore the computation time increases a lot. The CGAL algorithm is much slower but it's almost constant in time. In the end it's more stable.

Thank you for the explanation.
As you said, this issue occurs rarely. So we prefer DgmOctree structure because it is much faster than CGAL. Meanwhile, we will try to find a reasonable way to solve this problem before next update.

Hello, Daniel.
We made some important updates to the plugin. The main change is that we don't use any ANN libs in the raster and distance calculation steps, simpler structure and strategy are adopted instead. The speed should be increased. It has been submitted to github. Please compile and test it.
If the new plugin works well, please update the link http://www.cloudcompare.org/release/Clo ... bin_x64.7z. And please add CSF plugin to the part of "Newest plugins:" of the link http://www.cloudcompare.org/release/index.html.
Thank you.

Ok, that's done.

Hello, Daniel.
We just updated plugin to solve a bug of leaving some blank areas in the ground points. Please compile it and update the download link again. Sorry for inconvenience.
By the way, "CloudCompare 2.8 beta 7z archive version" on http://www.cloudcompare.org/release/index.html still linked to http://www.cloudcompare.org/release/Clo ... bin_x64.7z.

Thanks for the feedback, I've updated the 'release' page of the website.

You'll have to wait a little bit for the updated 2.8 beta version as I'm travelling for a few days.