Forum

Author Topic: Poor model details from good dense cloud  (Read 12885 times)

Bruno Andrieu

  • Jr. Member
  • **
  • Posts: 51
    • View Profile
Re: Poor model details from good dense cloud
« Reply #15 on: January 24, 2014, 09:40:23 PM »
here the model .....

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Poor model details from good dense cloud
« Reply #16 on: January 25, 2014, 12:33:39 PM »
Hi Bruno, as i se you have 12 cams, try to shoot same dataset with at least 24 cams and report back.....
----------------
www.mhb.sk

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14813
    • View Profile
Re: Poor model details from good dense cloud
« Reply #17 on: January 25, 2014, 09:00:35 PM »
Hello cbnewham,

Thank you. Will download and check it.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14813
    • View Profile
Re: Poor model details from good dense cloud
« Reply #18 on: January 27, 2014, 06:04:17 PM »
Hello cbnewham,

We're currently looking at the project file and it seems that some cameras are wrongly aligned (for example, image *3816.jpg).
Another comment is that there are a lot of camera calibration groups with the intermediate focal length positions for zoom-lens, since I'm not sure if all of them are really related to the exactly the same focal length (and not just approximate values stored in EXIF) I recommend to split all calibration groups (and also input initial parameters like focal length and sensor pixel size (both in mm) for every calibration group prior to splitting, since I also see that there are some NC cameras.
Best regards,
Alexey Pasumansky,
Agisoft LLC

cbnewham

  • Full Member
  • ***
  • Posts: 121
    • View Profile
Re: Poor model details from good dense cloud
« Reply #19 on: January 27, 2014, 06:32:12 PM »
Ok, thanks for the feedback so far. Although I am curious as to why that would impact on generating the model incorrectly from the dense cloud (as that's already bee created at that point).


Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14813
    • View Profile
Re: Poor model details from good dense cloud
« Reply #20 on: January 28, 2014, 11:25:15 AM »
Yes, it should not affect the mesh generation since the dense cloud has been reconstructed, but it could cause texturing problems.

However, we have been able to reproduce the issue with missing crown parts in 1.0.1.

Version 1.0.0 (build 1795) has been using different parameter values for meshing method, than those currently implied in 1.0.1. We are always trying to optimize and adjust the algorithms to get better output results. And even in 1.0.0 some thin model elements were reconstructed better, overall geometry has been very noisy and quite often had unwanted artifacts when higher quality settings were used. These problems has been reported on our forum:
http://www.agisoft.ru/forum/index.php?topic=1844.0
http://www.agisoft.ru/forum/index.php?topic=1914.0
And actually led to the very pixelated texture or even completely black texture.

So will try to improve the method of mesh generation (also basing on the project file you've provided), but in the current version there really could be problems with reconstruction of thin model elements.
Best regards,
Alexey Pasumansky,
Agisoft LLC

cbnewham

  • Full Member
  • ***
  • Posts: 121
    • View Profile
Re: Poor model details from good dense cloud
« Reply #21 on: January 28, 2014, 12:25:42 PM »
Thanks Alexey.

Do you have any plans to provide a UI control for the parameters so we can tweak them? Maybe even a manual way to specify thin parts of the cloud that need different params to the rest?

chadfx

  • Full Member
  • ***
  • Posts: 205
    • View Profile
Re: Poor model details from good dense cloud
« Reply #22 on: January 28, 2014, 08:06:49 PM »
+1 for mesh building parameters

It seems like you have been struggling to find a 'one size fits all' method for building the mesh and texture after the nicely detailed (and editable!) dense point cloud. I am working on objects from 2cm to 20m of all shapes and levels of detail. Photoscan continues to amaze me in its ability to handle so many variations and with such flexibility. But perhaps there could be a few more variables that we can adjust on our end to help fine tune the reconstructions.

Maybe there are simply too many variations in the types of reconstructions everyone is creating? It would be great to have some control over how the software decides to build contours and edges, the amount/type of interpolation (beyond the three limited & sometimes unpredictable settings). It would be great to have some options like angular thresholds, etc. Sometimes I really don't know what points to edit out of my dense clouds to get it to stop creating false edges/contours, but setting the mesh to no interpolation leaves simply too many holes.

Cheers, -Chad