[Solved] To many scans? Where is the border for crashing?

[AlexNoir]

Hello!

I'm working now since two days with David 3.8-3.10 (since the x64 Version is avaiable) on a complicate object where I made 82 scans. I allready reduced the the scans two times to 80% ( The last step was from the whole size of all scans from 2,3 to 1,6Gb - I'm in this moment not anymore happy about the quality, but I need the model!) but it's allways crashing when I want to proceed the global registration or the fusion of the scans. If it would be possible I would proceed the model without reducing the quality - my computer is good enought - but it seams like that the software is not really able to manage RAM over 4,5Gb - so what shall I do? It's mostly the same if I want to proceed BIG Objects, or really complicate objects with many scans. I'm not able to fuse because of the break down of the software. Is there still a borderline despite the x64 Version?

Please help me - it's important!

Alex

[ShadowTail]

I would suggest that you only load part of the scans into David, align those and save them.

Then load another bunch of scans and align them to one of the previously aligned scans.

That way you should be able to get everything lined up nicely while still staying within the capabilities of the program.

[AlexNoir]

Hello!

That is that what I'm actually doing - that's why I'm able to come to the Point so far where I am now. But this doesn't solve the Problem with the Fusion, becausse there I Need all scans together (especially because of the texture).

Alex

[MagWeb]

Hmm, until now I'd never run into this problem. But I'm working with textures not that often. Without textures you are not near to 2GB with such a number of partial scans (BTW: what's your cam's resolution?)

Are you sure to need all scans? There should be plenty of redundant surfaces which may be needed for automatic FreeAlign but which do not make the fusion result any better. First you should crop all useless stuff after a rough (free) alignment. After that cleaning save all partial scans just to be sure not to lose some work.
Fine and Global do not need that much overlaps. Maybe that's enough to solve your problems?

I could reproduce your crashes on Global alignment and Fusion with a similiar load of textured meshes:
Also here DAVID on WIN7 64 (Home Premium should handle 16GB) DAVID does not use memory above 4GB which causes a crash here too. Here the limit seems to be around 1.4 GB mesh load in total.
Other apps (also Lars' app) use the full available memory.

[MagWeb]

seems global fine registration has problems with more than 64 scans.

A reproduceable crash here:
-load a scan ( with or without texture)
-duplicate this scan until you've got 64 scans
-run run Global : Fine

-add one duplicate more
-run Global : Crash

Of course this does not make much sense, but this crash above 64 happens also with different meshes, no matter if aligned or not.

[MagWeb]

Double checked it with real scans (no duplicates):

DAVID crashes at a number of visible scans higher than 64 scans on my 64bit system running Global.
- this isn't related to the filesizes (tried also really small scans)

[MagWeb]

Sounds like some array(8,8) problem for Global?
Instead Fusion crashing isn't related to this 64 scans (it seems to be a pure memory issue) . While Fusion DAVID uses available memory resources.

[Sven]

Thanks, Gunter.
I can reproduce the Global Alignment problem. 64 scans are OK. 65 scans -> Crash.
We are looking into it.
There definitely is not a deliberate limitation to 64 scans.
Sven

[MagWeb]

Glad you could catch it - needed some time to find that borderline.

If there's some negative side effect solving this issue (e.g. slow down even doing global on only a few scans):
I'd leave it the current way and restrict Global to not more than 64 and give a message warning to the user. This is a pretty high number of scans and there's a way to handle bigger amounts of scans grouping all above 64 to scan_64 (which works fine).

BTW is there a "fix_scan" = a scan not being moved at all in Global - maybe scan_01? If yes, this could be a "bridge" between two "mezzo-global" runs.

[Sven]

hi Gunter,
No there is no "fix scan". Instead there is a function that prevents drifting, i.e. the CENTER POSITION of all scans remains stable.
We still haven't had time to fix this bug...
Sven

[Sven]

Markus has found and fixed the bug!
Each scan must be prepared for alignment, which we do in parallel. Seems we can only create 64 parallel threads!
The fix will be released with the next version, 3.10.

Thanks for reporting and investigating!

Sven

[AlexNoir]

Heey!

Sorrry that I ansswer only now - but I was away for work that lonng outside of my Country - And I didn't espect such a cool Report from someone else - thanks a lot therefore!

But I would have to add something: "there's a way to handle bigger amounts of scans grouping all above 64 to scan_64 (which works fine)." - is not a Option when it Comes to Texture - When I'm doing this, and saving the "new" Scan - the texture doesn't work anymore. So will the solution of using all scans be included in the new Release, or will there the same Problem happen, when I'm using more then 64 Scans? And yes These are a lot of scans, but there are many bigger objects waiting to the use of DAVID on them

Thanks again!

Alex

[Sven]

Hi Alex,

You are right, saving a GROUP of textured scans into one OBJ file --> the texture is lost.
You should group only to prevent the GFR bug, then ungroup before saving.

Of course, in the fix for the next version, DAVID can align more than 64 scans, and grouping won't be necessary any more.

Sven

[MagWeb]

Glad to hear about the fix!

Further detailed instruction to what Sven said about grouping and the suggested workarond:
If there are more than 64 scans split your Global alignment into two steps:
Combine scans belonging to the lower half of your object to get not more than 64 entries in the list (e.g. 63 scans + one group). Do a global alignment. The group is handled as being a single scan. Do not save the group but uncombine it after that global step. Now group some of the upper scans to get 63 scans + one, now different group again. Do a global alignment again. Ungroup after that again and do a fusion.
This workflow should keep the textures.