Forum

Author Topic: unwanted vertices inside the mesh  (Read 8619 times)

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14818
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #15 on: June 08, 2019, 06:05:55 PM »
Hello Mak,

Yes. Additionally you can also use the following tweak to eliminate the "bubbles" inside or under the surface:
Code: [Select]
main/mesh_visibility_trimming_radius
Recommended value is 10. By default the option is disabled (i.e. set to 0).
Best regards,
Alexey Pasumansky,
Agisoft LLC

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #16 on: June 08, 2019, 06:11:44 PM »
Alexey,
Thanks, I will test it out later tonight.

Cheers

Mak

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #17 on: June 08, 2019, 10:19:13 PM »
Hello Mak,

Yes. Additionally you can also use the following tweak to eliminate the "bubbles" inside or under the surface:
Code: [Select]
main/mesh_visibility_trimming_radius
Recommended value is 10. By default the option is disabled (i.e. set to 0).

Alexey,

Just started running some tests (first with main/mesh_visibility_trimming_radius 0) and I'm unfortunately getting exactly the same bad results as with 1.5.2: holes in the meshes, seam-line artifacts etc..

Edit: mesh_visibility_trimming_radius xx is like using gradual selection of connected component to delete unwanted polys/vertices but without any flexibility. If 10 is not enough or too much then you have to re-process the mesh..also if the "bubble" is connected to the main mesh, trimming it creates a hole !...

15.1:
https://imgur.com/eeocTWU

15.3:
https://imgur.com/4Ndl0Rf


To say that I'm disappointed would be gross understatement. Especially when 1.5.1 didn't suffer from those particular issues and meshes even had more detail at the same settings when using the Depth maps mesh generation workflow.
I'll repeat what I've been saying since March but the "Improved filtering of coarse depth map levels & Improved noise suppression for depth map based mesh generation" introduced in 1.5.2 are a step backward & 1.5.3 doesn't seem to have fixed any of this.

Back to 1.5.1 for me unfortunately.

Mak
« Last Edit: June 09, 2019, 02:31:43 AM by Mak11 »

Paulo

  • Hero Member
  • *****
  • Posts: 1303
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #18 on: June 09, 2019, 02:22:26 AM »
Dear support,

I would respectfully suggest that you prepare a small document explaining the main tweaks and their use....


They seem to be very useful but seem to be a mystery that is uncovered little by little....

A list of most useful tweak with their effect would greatly help!
Best Regards,
Paul Pelletier,
Surveyor

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #19 on: June 09, 2019, 02:00:53 PM »
Dear support,

I would respectfully suggest that you prepare a small document explaining the main tweaks and their use....


They seem to be very useful but seem to be a mystery that is uncovered little by little....

A list of most useful tweak with their effect would greatly help!

More importantly it would be great to not introduce under-the-hood & undocumented optimizations that either make things slower to process compared to the previous version or break things appart with no tangible advantages. This has been the case for a while now.  :(

Performance optimization where done to camera alignment when using Generic Pre-selection: it's faster which is great but a lot less reliable than before which results in the need to align with pre-selection disabled more often thus rendering this feature optimization somewhat moot if said feature can't be used with confidence anymore.

Depth Maps generation has been changed under the hood between 1.5.0 & 1.5.1 (to improve the Dense Cloud quality) which makes it 30% slower on average with no particular benefits when using the new depth maps meshing workflow (there's an undocumented tweak to revert to the old method -> "method main/depth_build_lod - FALSE") which was useful up until 1.5.2 was released.

And now two new Depth Maps optimization have been introduced in 1.5.2 which makes things even worse: it's slightly slower (you have to set "main/depth_build_lod -TRUE" or things will be even worst now) and the meshes generated are simply messed up in terms of topology with holes and unwanted polys all over the place.

For those using Metashape and generating meshes using the Depth Maps workflow sticking to 1.5.1 is IMO the best solution for now.

Mak
« Last Edit: June 10, 2019, 02:12:01 AM by Mak11 »

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #20 on: June 10, 2019, 01:50:07 AM »
To be perfectly clear and prove that something is wrong I just processed 2 of Agisoft's samples (Doll and Monument which anybody can download https://www.agisoft.com/downloads/sample-data/) using the depths maps workflow on Ultra High in 1.5.1 & 1.5.3 without changing any settings.

The results speak for themselves:

Doll

1.5.1: https://imgur.com/TS9auiR

1.5.3: https://imgur.com/Okkuc0Z

Monument

1.5.1: https://imgur.com/0zF6VlA

1.5.3: https://imgur.com/d49aMFb


Mak
« Last Edit: June 10, 2019, 02:12:54 AM by Mak11 »

zagnimarco1

  • Newbie
  • *
  • Posts: 42
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #21 on: June 16, 2019, 12:32:50 AM »
I'll try as soon as  possible

zagnimarco1

  • Newbie
  • *
  • Posts: 42
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #22 on: June 16, 2019, 04:20:31 PM »
I'm testing the mesh who had isolated parts inside,  at the moment seems works fine.

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #23 on: August 14, 2019, 06:31:43 PM »
To be perfectly clear and prove that something is wrong I just processed 2 of Agisoft's samples (Doll and Monument which anybody can download https://www.agisoft.com/downloads/sample-data/) using the depths maps workflow on Ultra High in 1.5.1 & 1.5.3 without changing any settings.

The results speak for themselves:

Doll

1.5.1: https://imgur.com/TS9auiR

1.5.3: https://imgur.com/Okkuc0Z

Monument

1.5.1: https://imgur.com/0zF6VlA

1.5.3: https://imgur.com/d49aMFb


Mak

2 months later....Metashape 1.5.4 ( & 6 Months after 1.5.1)..and Mesh generation via Depth Maps is still broken/useless:



Mak
« Last Edit: August 14, 2019, 06:34:15 PM by Mak11 »

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #24 on: August 14, 2019, 07:21:00 PM »
1.5.1:


1.5.4:


What a Joke  >:(

Sticking to 1.5.1 and then definitely moving to Reality Capture.

Mak

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14818
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #25 on: August 16, 2019, 09:42:14 PM »
Hello Mak,

Do you have any non-default tweaks set up in version 1.5.4? Like mesh_trimming_radius or mesh_visibility_trimming_radius, for example? If so, then what are the assigned values for them?

I have generated the model from the Doll.psz project in the release 1.5.4 version without any tweaks applied using High quality depth maps and Mild filtering (no volumetric masks applied) and got the following result, then rebuilt the mesh from the same depth maps with mesh_trimming_radius set to 30.

Best regards,
Alexey Pasumansky,
Agisoft LLC

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #26 on: August 17, 2019, 10:01:55 AM »
Alexey,
No tweaks where applied in both 1.5.1 d 1.5.4. Doll was generated at Ultra-High while the concrete slab was generated on High.

Cheers

Mak

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #27 on: August 19, 2019, 07:45:29 PM »
Hello Mak,

Do you have any non-default tweaks set up in version 1.5.4? Like mesh_trimming_radius or mesh_visibility_trimming_radius, for example? If so, then what are the assigned values for them?

I have generated the model from the Doll.psz project in the release 1.5.4 version without any tweaks applied using High quality depth maps and Mild filtering (no volumetric masks applied) and got the following result with mesh_trimming_radius set to 30.

Doll.PSZ in 1.5.4
High quality depth maps and Mild filtering (no volumetric masks applied) and got the following result, then rebuilt the mesh from the same depth maps with mesh_trimming_radius set to 30:

Still unuseable(!)  ::) :



If there's (obviously) no plan to fix this/revert back to a functional version of the Depth Maps based meshing workflow then it would be great to at least acknowledge it so we don't waste our money/time.

Thanks

Mak
« Last Edit: August 19, 2019, 07:48:21 PM by Mak11 »

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14818
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #28 on: August 19, 2019, 09:18:17 PM »
Hello Mak,

The current depth maps based mesh generation method is still being developed and optimized, including the results of reported problems studying.

Unfortunately, there is not a kind of internal parameter value that can be changed and that will fix everything. Since version 1.5.1 the algorithm has been already revised as there were less obvious issues that were fixed.
At the moment the main development is focused on out-of-core implementation of this method in order to allow achieving results even with moderate amount of RAM (currently some projects in Ultra quality may require a few hundreds of GB RAM to complete). But at the same time all reported issues are also investigated.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Mak11

  • Sr. Member
  • ****
  • Posts: 374
    • View Profile
Re: unwanted vertices inside the mesh
« Reply #29 on: August 19, 2019, 09:38:48 PM »
Hello Mak,

The current depth maps based mesh generation method is still being developed and optimized, including the results of reported problems studying.

Unfortunately, there is not a kind of internal parameter value that can be changed and that will fix everything. Since version 1.5.1 the algorithm has been already revised as there were less obvious issues that were fixed.
At the moment the main development is focused on out-of-core implementation of this method in order to allow achieving results even with moderate amount of RAM (currently some projects in Ultra quality may require a few hundreds of GB RAM to complete). But at the same time all reported issues are also investigated.

Alexey,

There are indeed some instances where the current algorithm gives better results that the one in 1.5.0/1 but those improvements are vastly out-weighted by the documented drawbacks which are unwanted vertices all over the place and holes (playing around with undocumented tweaks can be a major time waste).
Focusing on being fully out-of-core is great (and needed!) but the quality of the mesh is still IMO the most important part.  1.5.1 was still far from perfect on this front but still more usable than the current implémentation. I'm always available to beta test or provide samples if needed.

Mak