M3C2

[ilovecoffee1125]

The M3C2 algorithm will not run. I am comparing two river banks. I set the parameters to: diameter: 1.0, projection: .2, max depth: .5 and then use the multi scale tab under the normals tab, Min: .4, Step: .4, Max: 1.6. I set the registration error to usually between .01-.05. depending on what my error was upon exporting from another program. Every time I try to run the comparison is says "not responding". It was working great all morning and then early afternoon it stopped. I used all the same settings as stated above.

Thank you.

[daniel]

Isn't it an issue with the actual units of the cloud? If the neighborhoods are very big it can take a long time.

And beware that computations are done in parallel, using all cores if possible. Therefore the system may be blocked for a while (this is why Windows says 'not responding' but it doesn't mean that CC has crashed). You may have to wait a little to see something moving. And it depends on what your computer / Windows is doing in parallel of course (this may explain the differences with the behavior you observed in the morning).

[ilovecoffee1125]

Daniel,

When you say 'units of the cloud' could you be specific? Are the parameters I have set (listed in the previous post) reasonable for running comparisons? Would you suggest changing any? Our goal is to detect small changes in erosion/deposition along bank shores.

Thank you.

[daniel]

I mean if your units are in millimeters and the scales you inject in meter for intsance? Which would end in way too big neighborhood extraction volumes?

But that's just a possibility. The other points are also plausible causes. And last but not least don't hesitate to send me the file if you want me to test (cloudcompare [at] danielgm.net).

[ilovecoffee1125]

Daniel,

The files I am working with are very large. A billion points some of them. And our scans are in meters. Is this why the M3C2 is not working?

[ilovecoffee1125]

Daniel,

I have sent you a link to some of my scans. I tried to run the M3C2 again last night and it would not work. Let me know what you think of the parameters I attached.

Thank you.
Carly

[daniel]

Thanks, I'll look at them asap.

[daniel]

For the records, here is (part) of my email answer:

I think you are facing too problems at the same time:
- first there are quite a lot of points, so you can't use all of them as 'core' points (unless you have a monster PC). The principle of 'core' points is to use less points than the actual number of points in very dense TLS scans in order to cope with the complexity of the M3C2 computation. The default subsampling value suggested by M3C2 is way to high (this is actually caused by the second issue below) but you can start with 0.1 meters and then reduce this radius later when you are happy with the settings. For a first run you should keep less than 1 million core points.

- second, the two clouds have a lot of scattered points, quite isolated and very far from the others. These points add a lot of complexity to the spatial analysis while they are (apparently) not interesting. So you'd better remove them first with a quick interactive segmentation step (use the 'scissors' tool to keep only the points inside a 2D polygon - see http://www.cloudcompare.org/doc/wiki/in ... ation_Tool)

[ilovecoffee1125]

Simple question: what does M3C2 stand for?

CS

[daniel]

I've updated the link to the original article on the wiki.

And the answer is... "Multiscale Model to Model Cloud Comparison" :D