Agisoft Metashape

Agisoft Metashape => Bug Reports => Topic started by: Kjellis85 on March 04, 2020, 01:18:26 PM

Title: Broken high res mesh
Post by: Kjellis85 on March 04, 2020, 01:18:26 PM
EDIT: ran it again with 100M face count (original set at 0) and it worked. Could be an issue with 0 face count.

Tried building mesh from a 154M pointcloud (generated in MS) but get this weird looking mesh, that is supposedly 373M poly high. Have made this previously with no problem.

2020-03-04 08:32:40 BuildModel: quality = Ultra high, depth filtering = Mild, source data = Dense cloud, surface type = Arbitrary, face count = 0, interpolation = Enabled, vertex colors = 1
2020-03-04 08:32:40 Generating mesh...
2020-03-04 08:32:40 Grid size: 8990 x 1574 x 22341
2020-03-04 08:32:40 Tree depth: 15
2020-03-04 08:43:37 Tree set in 656.832s (154227843 points)
2020-03-04 08:43:51 Leaves/Nodes: 1023032137/1169179585
2020-03-04 08:48:17 Laplacian constraints set in 265.912s
2020-03-04 08:48:20 Depth[0/15]: 1
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0 / 0
2020-03-04 08:48:20 Depth[1/15]: 8
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0 / 0
2020-03-04 08:48:20 Depth[2/15]: 64
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0.00100017 / 0.000999928
2020-03-04 08:48:20 Depth[3/15]: 512
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0.000999928 / 0.00200009
2020-03-04 08:48:20 Depth[4/15]: 4096
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0.00499988 / 0.00600004
2020-03-04 08:48:20 Depth[5/15]: 32768
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0.0219996 / 0.0180004
2020-03-04 08:48:20 Depth[6/15]: 13472
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0.00800037 / 0.0269995
2020-03-04 08:48:20 Depth[7/15]: 36768
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0.0190001 / 0.0489998
2020-03-04 08:48:20 Depth[8/15]: 121864
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0.0429997 / 0.0740004
2020-03-04 08:48:20 Depth[9/15]: 441448
2020-03-04 08:48:20    Evaluated / Got / Solved in: 0 / 0.136 / 0.174
2020-03-04 08:48:20 Depth[10/15]: 1682024
2020-03-04 08:48:22    Evaluated / Got / Solved in: 0 / 0.701004 / 0.821996
2020-03-04 08:48:22 Depth[11/15]: 6582408
2020-03-04 08:48:26    Evaluated / Got / Solved in: 0 / 2.021 / 2.387
2020-03-04 08:48:26 Depth[12/15]: 25565256
2020-03-04 08:48:43    Evaluated / Got / Solved in: 0 / 7.639 / 8.752
2020-03-04 08:48:43 Depth[13/15]: 95368864
2020-03-04 08:49:48    Evaluated / Got / Solved in: 0 / 29.501 / 32.825
2020-03-04 08:49:48 Depth[14/15]: 318797312
2020-03-04 08:53:35    Evaluated / Got / Solved in: 0 / 102.649 / 113.56
2020-03-04 08:53:35 Depth[15/15]: 720532720
2020-03-04 09:01:52    Evaluated / Got / Solved in: 0 / 234.363 / 236.971
2020-03-04 09:01:53 Linear system solved in 815.592s
2020-03-04 09:02:34 Got Iso-value in 41.031s
2020-03-04 09:02:34 Iso-Value -0.0517755
2020-03-04 09:16:43 373736974 faces extracted in 848.316s
2020-03-04 09:29:55 Calculating vertex colors...
2020-03-04 09:29:55 processing nodes...  done in 11.708 sec
2020-03-04 09:30:07 calculating colors...  done in 2177.03 sec
2020-03-04 10:06:25 Peak memory used: 79.40 GB at 2020-03-04 08:43:34
2020-03-04 10:06:25 Finished processing in 5625.24 sec (exit code 1)
>>>
Title: Re: Broken high res mesh
Post by: Alexey Pasumansky on March 04, 2020, 03:43:50 PM
Hello Kjellis85,

We'l try to reproduce the problem on similar size project, but if you have some spare processing time, can you please check if setting the face count to 360 millions, for example, gives you the same problem or not?

If you are able to open the exported high-poly  model in the external application, can you confirm that the issue is still observed (and it is not a rendering issue in Metashape)?
Title: Re: Broken high res mesh
Post by: Kjellis85 on March 04, 2020, 04:02:08 PM
I can run it at the end of the working day to check if it stays the same at 360M. I did a test uploading to sketchfab and the model had the same issue. I don't have any desktop software that supports such high poly models.
Title: Re: Broken high res mesh
Post by: Kjellis85 on March 06, 2020, 10:01:10 AM
Ran it again at target 360M and got the same faulty mesh as when I had the target set at 100M.
Title: Re: Broken high res mesh
Post by: Alexey Pasumansky on March 06, 2020, 06:09:13 PM
Hello Kjellis85,

If you decimate the same mesh to the 10 million face count - does it appear to be correct in the Model view or the same mess of the polygons is observed?
Title: Re: Broken high res mesh
Post by: Kjellis85 on March 09, 2020, 11:15:15 AM
Decimated the broken 360M model to 10M and it appears to be in order, at least in MS model view. I realize that a workaroud could soleve the problem, by steering away from the large model, but it is not a solution for the long run.
Title: Re: Broken high res mesh
Post by: Alexey Pasumansky on March 09, 2020, 01:27:49 PM
Hello Kjellis85,

If the decimation "fixes" the problem, then it seems that the issue is related to the visualization in the Model view only.

Which GPU is used for the OpenGL rendering and which driver version you have installed?
Title: Re: Broken high res mesh
Post by: Kjellis85 on March 09, 2020, 01:36:22 PM
I am running NVIDIA RTX 2070 with driver 26.21.14.4128
Title: Re: Broken high res mesh
Post by: Alexey Pasumansky on March 09, 2020, 03:09:19 PM
Hello Kjellis85,

I can suggest to install latest NVIDIA drivers from NVIDIA web-site. If it doesn't help, please try to enable VBO in Metashape advanced preferences and set VBO limit to 4096 MB.
Title: Re: Broken high res mesh
Post by: Kjellis85 on March 09, 2020, 03:14:27 PM
I have VBO activated and set limit to 41280 MB(!). Could that be an issue, that it has too much memory allocated? I was under the impression that the more the better was the general rule for VBO?
Title: Re: Broken high res mesh
Post by: Alexey Pasumansky on March 09, 2020, 03:15:09 PM
Hello Kjellis85,

VBO limit shouldn't exceed available VRAM.
Title: Re: Broken high res mesh
Post by: Kjellis85 on March 09, 2020, 03:16:46 PM
Ok, I will change to a lower limit, within my Vram. Will run another test for tomorrow.