Export ANY Point Cloud

[Lennart Hildebrandt]

Hello Community,

currently I am trying to export either the sparse or the dense point cloud from my model to import it in CloudCompare. 
Theirfore I use the nodes "ExportColoredPointCloud" + "ConvertSfMFormat". 
In the Meshroom Manual for the Input of SfM Data is written  > SfMData file containing a complete SfM. < 

But the pipeline runs always into an error no matter if I connect the node with "StructureFromMotion" or "Meshing". In the log I do not get any error notification..

Does anyone has an idea or knows how to properly export a point cloud from Meshroom.

Regards Lennart

[Theo Hartsook]

Hi Lennart,

This is the method I use to export a point cloud:

1. Texturing node (Dense SfMData) connected to ExportColoredPointCloud (Input SfMData)
2. ExportColoredPointCloud (Input SfMData) connected to ConvertSfMFormat (Input)
3. On the ConvertSfMFormat node, I set Describer Types to unknown so that the points are exported.

I attached a screenshot to show how my nodes look. Hopefully this is some help. 

Best regards,

Theo

[Lennart Hildebrandt]

Hi Theo,

thanks for your answer. Trying your solution I run into an error again.

The following is what the Log says
[warning] Alembic LOAD: NO OBSERVATIONS_WITH_FEATURES: , mvg_visibilityFeatId: 0, mvg_visibilityFeatPos: 0, OBSERVATIONS_WITH_FEATURES flag: true 

Kind regards

[Theo Hartsook]

Hi Lennart,

That warning indicates that empty SfM data is passed to the triggering node. Unfortunately there are several reasons that could be happening, so it is a little difficult to diagnose the problem. I can give some troubleshooting suggestions from my own experience, and maybe someone else in the community knows some more:

1) pipeline is misconfigured: if at any point in the pipeline a SfMData input is incorrect or empty, that warning could trigger (since there are no points to load).
2) failed reconstruction: if the reconstruction failed, then any nodes depending on it will load empty data.
3) reconstruction corrupted: It's possible that the SfMData has been corrupted by some external process. This one happened to me recently due to a drive failure.

To be totally honest, I usually do not spend so much time troubleshooting that type of warning, unless I really want to keep a particular project. Maybe 30-60 minutes. If that's not enough I usually recreate the project from scratch instead of trying to find what in particular went wrong.

Best regards,
Theo

[Lennart Hildebrandt]

Hi Theo,


first: thanks a lot for your time and help!

For my current project it is not necessary to get the point cloud. Would be more a nice-to-have thing.
For me it it more important to find out how to receive a point cloud from the pipeline for the future. And as you showed me a way that works (at least for you) I will keep that in mind.
Right now I receive a mesh I can work with so I am not sure if your points will actually help/fit with my project.
But since I am still trying to figure out how Meshroom works I am producing projects over projects anyway. 


BTW..do you have/use a documentation about all the nodes yourself or do you work with just your knowledge?
I am missing so many explanations in the official documentation and in Meshroom as well so it is hard to understand the algorithms and working processes behind the software.


Kind regards,
Lennart

[Theo Hartsook]

Hi Lennart,

I apologize for the late response. I generally use the documentation and my own knowledge as a starting place, but I find so much helpful information on the GitHub page: https://github.com/alicevision/meshroom

I agree the documentation is a little challenging. I think it assumes a lot of computer vision knowledge compared to other SfM software. I have some background in computer vision and SfM so lately I've been wondering if I could add some extra explanations and references for some of the more complex nodes (just for the ones I believe I understand well, such as FeatureExtraction). I haven't contributed to a project before but maybe elaborating on some nodes in this group could be a good start!

Best regards,

Theo