Anybody use general selection to clean up sparse points?

[SB]

Hello

I've been looking at a lot of Python scripts and NONE of them try to clean up the point clouds before generating a mesh.  Am I the only person who uses gradual selection to remove uncertain points?

I try to reduce reprojection error to 1.0 - 1.2 and remove uncertain points at a level of 400 for every chunk before generating the dense clouds.

Am I doing the correct procedure??

Thank you

[Heinrich]

I always do so; it is a key step for getting good models with very littel noise.
However, the exact parameters vary, thus I do not need the ability to skipt this (but it would be nice=)

[sc]

Would be real interesting to see a test case with and without these adjustments

[doctorben]

I've made a test case with gradual selection.
In my example project i have a chunk with 265 photos aligned (highest accuracy, point limits 0).
Result is 2.241.674 Tie Points.

Then for the next step i've duplicated the project and put an gradual select in one of these:
Reprojection error 0.5
Reconstruction uncertainty 50
Image count 3
Projection accuracy 10
Result is 294.788 Tie Points.

Thereafter, in each project the dense cloud is generated with high quality.

Here is the comparsion with each photo/project (first screenshot is always the project without gradual selection)

Sparse Cloud:
http://imageshack.com/a/img901/7139/ygupUY.png
http://imageshack.com/a/img910/6843/okiUsy.png

Details:
http://imageshack.com/a/img538/1027/voCPok.png
http://imageshack.com/a/img673/9691/UMds92.png

http://imageshack.com/a/img673/1631/oFHd9e.png
http://imageshack.com/a/img673/6552/WcTy0Y.png

http://imageshack.com/a/img909/698/yutlRk.png
http://imageshack.com/a/img909/185/kqmEup.png

http://imageshack.com/a/img538/5281/2sZC3I.png
http://imageshack.com/a/img911/179/AfjxZU.png

In my opinion theres is no significant better results in dense cloud, only a little bit less noice.
Maybe there can be other results with limited tie points.

[sc]

Thanks for taking the time doing this In the detail shots you can see that you get less 'background points' sticking to the outline of the silhouettes. But its not consistent and doesn't happen everywhere.

[SB]

Thanks for the test.  Even if it only reduces the noise a little it's worth using to me. 

[Mohammed]

Hi doctorben

 did you optimize sparse point cloud after you clean it?
thanks for shown us ur comparison!

[doctorben]

Yes, i have done the optimization after cleaning up the sparse cloud.