Forum

Author Topic: Poor model details from good dense cloud  (Read 12289 times)

cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Poor model details from good dense cloud
« on: January 24, 2014, 12:32:17 AM »
I would like to know why the model becomes so degenerate even though the dense cloud has ample points and clearly defines the surface. There appears to be no way to adjust how the dense cloud to surface algorithm works.

Note the stems disappear on the crown and the curved edge nearby has a jagged section taken out of it. Strangely, I've built this one before (but with a few less cameras - not, I should add, imaging this particular area. I added them to the current version for detail elsewhere) and the stems were generated and the edge was not jagged. So something is either wrong with the build or there is some magical change that happens when more cameras were added.

This was with the latest build (1812).



cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Re: Poor model details from good dense cloud
« Reply #1 on: January 24, 2014, 01:03:47 AM »
Here is the earlier one I built back on the 8th of January. This had less cameras overall, but exactly the same cameras with this area in view as on the later reconstruction above.

Note that the dense cloud has only 16 million points in this example, and 20 million on the later one above.

Why the difference in quality of the model?!


Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Poor model details from good dense cloud
« Reply #2 on: January 24, 2014, 08:47:49 AM »
how much triangles has the mesh ? from what i see it has too liittle trias......
----------------
www.mhb.sk

cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Re: Poor model details from good dense cloud
« Reply #3 on: January 24, 2014, 10:23:01 AM »
The number of triangles makes no difference. I always have it set to about 500,000 and increasing them leaves it in the same state - only with more triangles... :-/

Anyway, overnight I re-ran the earlier set of photos that came out ok and it now looks like the later one with missing details and broken edges - so it looks like something has changed in the new build of PS.

Point cloud: 700,000 points
Dense: medium, moderate filter
Mesh: 600,000
Decimate: 500,000

I guess this problem stems from solving the texture problem of the previous build? If so, that's a shame. I'm reverting back and will run this through again. If it works I'll pass on this build - I'd rather work around the texture problem than have a mesh which is clearly not generating properly.
« Last Edit: January 24, 2014, 10:45:12 AM by cbnewham »

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Poor model details from good dense cloud
« Reply #4 on: January 24, 2014, 10:38:50 AM »
cbnewham, just set triangle count to 5 milions, reconstruct from the Pcloud that you already have and report back.......

from this small screenshot its hard to say proper triangle count, but i see that PCloud is dense so the details should be reconstructed..... its better to have mesh reconstructed with higher triangle count as you CAN later decimate to the optimal value.......
----------------
www.mhb.sk

cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Re: Poor model details from good dense cloud
« Reply #5 on: January 24, 2014, 10:49:08 AM »
I set the triangle count at one point to several million and still ended up with everything the same, except with far more triangles in the broken pieces.

I will try this again to double check, but in the older example that came out fine, I can tell you that my mesh will have had no more than 1 million before decimation.

I am still suspicious about build 1812...  :-)


cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Re: Poor model details from good dense cloud
« Reply #6 on: January 24, 2014, 12:03:27 PM »
It is as I said - lots more triangles but a degenerate surface.

This was with 5 million triangles. The surface is just not being generated correctly from the dense cloud. My next test will be with the earlier release, and I suspect it will produce a much nicer result judging from my results on the 8th of Jan.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14107
    • View Profile
Re: Poor model details from good dense cloud
« Reply #7 on: January 24, 2014, 01:48:10 PM »
Hello cbnewham,

Are you sure that you are generating mesh model based on dense cloud and not on sparse?

Maybe you can share the project file with us?
Best regards,
Alexey Pasumansky,
Agisoft LLC

cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Re: Poor model details from good dense cloud
« Reply #8 on: January 24, 2014, 01:52:20 PM »
Hi Alexey,

Yes, this is with the dense cloud.

Once the generation with the older PS release is finished and I can see what has happened, I'll send you the project. What is the best way to do this? It's pretty big.

cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Re: Poor model details from good dense cloud
« Reply #9 on: January 24, 2014, 06:06:53 PM »
I've now confirmed this.

Build 1795 produces the correct model (values for the various phases as given previously, although I made the mesh 700,000 and didn't decimate to 500,000).




Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14107
    • View Profile
Re: Poor model details from good dense cloud
« Reply #10 on: January 24, 2014, 06:16:29 PM »
So, just to be sure, have you used the same dense cloud for tests in both versions?
Best regards,
Alexey Pasumansky,
Agisoft LLC

cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Re: Poor model details from good dense cloud
« Reply #11 on: January 24, 2014, 06:38:53 PM »
Alexey,

I took the Project as created in 1795, reinstalled 1812 and loaded it.

The model looks fine (as generated in 1795).

I then regenerated the model from the dense point cloud with 700,000 polys.

The model no longer looks fine - again missing parts and with bad edges.

So therefore the model generated by 1812 using the same point cloud as 1795 is no good.



Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14107
    • View Profile
Re: Poor model details from good dense cloud
« Reply #12 on: January 24, 2014, 06:40:46 PM »
Hello cbnewham,

That's what I wanted to know, thank. So it will be important for us to see the project file with the dense cloud.
Best regards,
Alexey Pasumansky,
Agisoft LLC

cbnewham

  • Full Member
  • ***
  • Posts: 120
    • View Profile
Re: Poor model details from good dense cloud
« Reply #13 on: January 24, 2014, 07:19:19 PM »
I've shared a dropbox folder with you via your

info@agisoft.ru

address.

It contains the model. (or will do - at 156 Mb and I'm on a slow link it will take a while to synch)


« Last Edit: January 24, 2014, 07:23:06 PM by cbnewham »

Bruno Andrieu

  • Jr. Member
  • **
  • Posts: 51
    • View Profile
Re: Poor model details from good dense cloud
« Reply #14 on: January 24, 2014, 09:39:23 PM »
Hello

I have a similar problem but much worse in my case.
The object is a piece of soil with wheat plants and some plastic elements.

The point cloud looks promising but plants and plastic elements nearly disappear when building the model...

Point clouds has 64 000 000 points
3D Model has 12 000 000 faces