Pb to convert a *.las file convert from a *.ptx file with CloudCompare

[Thomas62149]

Hi Markus,

I'm testing PotreeConverter 2.01 on a large series of .las files. It works great for a big majority of them but Impossible to convert this .las file with PotreeConvert 2.01. PotreeConverter seems to fail in the last "Indexing" operation (it write "0% INDEXING" for 30 second in the console and then close. The .ptx file is exported by leica software "Cyclone register 360" from a BLK scanning. Here is the link to download the .las file : "https://www.grosfichiers.com/u5mcnndZsPC". I've tried to direct convert the .ptx with PotreeConvert 2.01 but it crashes the .exe at the beginninbg of the analyse of the File.

Thanks for your answer

Best regard

Thomas MARTEL

[smcavoy12]

Hi Thomas,

I tested, it failed. I used LAStools lasduplicate64, removed 3080 duplicate points, then it ran fine.

I've been having having to do this with most of my TLS datasets now, it's now a standard part of my process, prepping files for potree. e57 to seperate las scans, run lasduplicate on the whole directory.

[midnight-dev]

Did this workload normally succeed without issue on 1.7? If this is more of a workaround than it is a workflow, then this could be a big red flag that there's been a regression in program behavior. To be fair, a new major version - especially a total rewrite - is expected to be different.

I believe the intended behavior is now to remove the data of confirmed duplicates, no matter the amount; however, I think it's only partially implemented. Need to revisit this after Markus confirms duplicates are gracefully handled.

[smcavoy12]

@midnight-dev on 1.7 it stalls up, stuck on "READING" for 20 minutes, then indexing the first 1 million. Very slow, I didn't let it go farther. On other files with small numbers of duplicates the behavior I've observed is MASSIVE slowdown, but eventual completion. I can't provide specific examples, but I'll save them from here on out if its useful to you.

[Thomas62149]

Thanks for your answers. The duplicates point are the problem. Markus is working on that type of problems to make PotreeConverter more robust. I use the command line "-SS SPATIAL 0.001" in cloud compare to sub sample the cloud and then the integration of the las in Potree works perfectly. Thanks

[m-schuetz]

I've made a build with a fix here: http://potree.org/temporary/PotreeConverter_drop_duplicates_2020.09.09.zip

If you had problems with ERROR due to duplicates, please give it a try and let me know if it works. Note that this fixes cases where there are tens of thousands of duplicates, perhaps even a couple of millions. It likely won't work if there are tens of millions of points at the exact same coordinate, in which case you'll still have to sanitize the data before using the converter.

[FrenchTrott]

Hi Markus, it worked on our end, on 3 problematic files.

Thanks !

When you talk about sanitizing a file, what's the best method you think?

[m-schuetz]

When you talk about sanitizing a file, what's the best method you think?

You could use Cloud Compare command line mode, as suggested by @Thomas62149

I use the command line "-SS SPATIAL 0.001" in cloud compare to sub sample the cloud and then the integration of the las in Potree works perfectly.

[FrenchTrott]

Yes but CC uses as much RAM as the file size .... If the file is 300 GB, it's complicated ... Do you have another idea?

[m-schuetz]

Do you only have a single large 300GB file? Aren't there tiled files or separate scan locations? If you only have the 300GB file, you might have to split it into smaller tiles, and apply the subsampling operation on each file. I'm not sure if Cloud Compare can partition a point cloud into tiles, but lastiles can.

[m-schuetz]

You could also give PDAL a try. Perhaps thinning might help. Generally, any subsampling/thinning operation might also drop points that were okay, but if you set the minimum distance/radius/spacing arguments reasonably low, there is a good chance that only duplicates or almost only duplicates will be removed.

[FrenchTrott]

Thanks Markus.

This last update, of this issue, will be integrated when?