Grouping Align Strategy

[MagWeb]

Hi out there,

after doing many scans from different directions and passing them over to Shapefusion you might end in a messy bunch of results which have to be sorted in Shapefusion. You might have lost your overview which one should be aligned to an otherr result? Don't worry, let DAVID do the job.

To make DAVID search for the best align target simply group all meshes but the one you search a target for:

Using the Free align tool, DAVID aligns the mesh to the best fitting member of the group:

[MagWeb]

Ungroup all meshes, go to the next mesh you want to find a target for, group all the rest and align the single mesh to the group using Free align. This time DAVID may decide another target mesh:

Don't worry, ungroup, go to the next mesh to be aligned, group the rest and align again:

Fortunatly DAVID decided this time to place it next to the previous one

... repeat ths group /ungroup workflow until there's no single mesh any more:

[MagWeb]

Now its a good idea to choose the mesh aglomeration with the fewest partial meshes.
Group the meshes of this aglomeration as Group A and all the rest as Group B. Align Group A to B:

Ungroup all.
Same procedure with the next aglomeration: Group all members, Group the rest Align A to B:

If there's no aglomeration left to align ungroup all and do GlobalAlign:

No need to think about which partial scan matches to an other mesh.

Hope that helps a bit

Gunter

[Simon]

Hi Gunter,
once again your ingenuity is very impressive! Thanks for sharing!

[MagWeb]

Hi Simon,

thanks for the flowers.
Think this could be automized - the only missing thing is a way to detect the aglomerations.

[Simon]

Think this could be automized

Yes it could. However, an adapted algorithm for multi-matching could run much faster. It would be intressting to see if these sheep scans can be aligned with Lars Multimatching and a long it takes.

BTW: Multi scan matching is still an important and interesting issue for us. I'm thinking about collecting a lot of scans which serve as database for testing new multi-matching algorithms.
@Gunter and @All: I would be glad if you can make some of your nice multi-part scans available for research and testing purposes.

[MagWeb]

Hi Simon,

Yes, Lars' app can align the sheep without any align errors.

Running the same set of sheep scans 10 times (confirming dialogs as soon as possible) with loaded scans I get :

1:24min from clicking Start to the final single blue result if Lars' app doesn`t call a confirmation (1 time of 10)
1:33min from clicking Start to the final single blue result if Lars' app calls a confirmation once ( 1time of 10)
1:54min (averaged) from clicking Start to the final single blue result if Lars' app calls a confirmation twice (8 times of 10)

EDIT: Some box to drop shared files?

[BoneIndustries]

I've found that it is easier to bring the scans into Meshlab for aligning and editing after scanning with David's.

The rest of Mister P's videos are pretty good also.
https://www.youtube.com/watch?v=4g9Hap4rX0k

[hal]

Good totorial about aligning, Gunter.
Thanks
Mattia

[NARMELLA]

Hi Gunter,

It has been and is a good idea to align scans correctly.
I have used and use your idea in many scans.

Here is a sample :

I scanned a cover of mirror in 3 main positions and aligned them separately .

Outside scans

Side scans

Inside scans

[NARMELLA]

Then i inserted all of them in to david session and used combine command to add them in 3 separate groups

And aligned them ( roughly )

And at last , I used uncombine command + Global Fine Registration

[NARMELLA]

And here is the result

Thank you for sharing your great idea

NARMELLA

[Ingo]

Thanks Gunter for your ideas,
To add to the discussion, I compared your strategy on a scan of a table tennis ball with alignment markers surrounding it.
I did 12 scans (30° each) on a turntable and cleaned the scans.
Afterwards I aligned 10 times the (same!) 12 scans with Gunters strategy.
As it turns out, the results are different for most of the different runs (see attached picture), although the starting point was the same each time.

The second version occured two times.
So, out of 10 times the correct alignment is found 3 times.
In all other cases different amounts of aglomerations can be found.
It seems, that the algorithm for alignment has different solutions each time it is started, even if the starting point is the same.
Over 12 alignments this adds up to a high diversity.

Best regards
Ingo

[NARMELLA]

Really great strategy

:

Thanks GUNTER

NARMELLA

[WalterMo]

Very good work, Narmella! I mean all what you have scanned and posted in the last time.
And specially thanks for your tutorial 'LET'S SCAN'!

It's a pity that Gunter isn't still working for this forum. Let's hope that he will return sometime.

Walter