RefineMesh Killed

[sjmduncan]

Describe the bug
RefineMesh gets 'Killed' before producing an output scene, and without a useful error message.
With -v 3 it produces some .ply files which appear to have undergone some refinement (Refine1.ply Refine2.ply, Refined1.ply, Refined2.ply).

All of the other parts of the pipeline seem to work fine, and I can produce a textured unrefined mesh with no problems.

OS is Ubuntu 18.04
Built using the script from this repository but with -DOpenMVS_USE_CUDA=OFF
Scene is imported from a COLMAP sparse reconstruction

The output looks like this:

> RefineMesh scene_dense_mesh.mvs --decimate 0.5 --max-face-area 16 -v 3
08:58:57 [App     ] Build date: Mar 22 2021, 08:38:37
08:58:57 [App     ] CPU: Intel(R) Core(TM) i5-6300U CPU @ 2.40GHz (4 cores)
08:58:57 [App     ] RAM: 7.65GB Physical Memory 0B Virtual Memory
08:58:57 [App     ] OS: Linux 5.4.0-67-generic (x86_64)
08:58:57 [App     ] SSE & AVX compatible CPU & OS detected
08:58:57 [App     ] Command line: scene_dense_mesh.mvs --decimate 0.5 --max-face-area 16 -v 3
08:58:57 [App     ] Scene loaded (437ms):
	4 images (4 calibrated) with a total of 76.15 MPixels (19.04 MPixels/image)
	0 points, 322398 vertices, 644269 faces
08:58:57 [App     ] Refine mesh at: 0.25 image scale
Decimated faces 322134 (100%, 2s988ms)           
08:59:01 [App     ] Mesh decimated: 644269 -> 322135 faces
08:59:01 [App     ] Removed 0 zero-area faces
08:59:01 [App     ] Removed 0 duplicate faces
08:59:01 [App     ] Removed 0 non-manifold faces
08:59:01 [App     ] Removed 0 degenerate vertices
08:59:01 [App     ] Removed 0 unreferenced vertices
08:59:02 [App     ] Removed 0 duplicate vertices
08:59:02 [App     ] Split 0 non-manifold vertices
08:59:02 [App     ] Closed 0 holes and added 0 new faces
08:59:02 [App     ] Cleaned mesh: 161331 vertices, 322135 faces (4s320ms)
08:59:02 [App     ] Removed 0 zero-area faces
08:59:02 [App     ] Removed 0 duplicate faces
08:59:02 [App     ] Removed 0 non-manifold faces
08:59:02 [App     ] Removed 0 degenerate vertices
08:59:02 [App     ] Removed 0 unreferenced vertices
08:59:02 [App     ] Removed 0 duplicate vertices
08:59:02 [App     ] Split 0 non-manifold vertices
08:59:03 [App     ] Closed 0 holes and added 0 new faces
08:59:03 [App     ] Removed 0 non-manifold faces
08:59:03 [App     ] Removed 0 non-manifold vertices
08:59:03 [App     ] Cleaned mesh: 161331 vertices, 322135 faces (713ms)
08:59:04 [App     ] Mesh imported: 164919 vertices, 329206 facets (618 border edges)
08:59:29 [App     ] Edge size in [0.00193519, 0.119355] (requested in [0.0149313, 0.119451]): 44018 ops, 50 iters
08:59:29 [App     ] Mesh exported: 139976 vertices, 278588 facets (1350 border edges)
08:59:30 [App     ] Removed 0 zero-area faces
08:59:30 [App     ] Removed 0 duplicate faces
08:59:30 [App     ] Removed 0 non-manifold faces
08:59:30 [App     ] Removed 0 degenerate vertices
08:59:30 [App     ] Removed 0 unreferenced vertices
08:59:30 [App     ] Removed 0 duplicate vertices
08:59:30 [App     ] Split 0 non-manifold vertices
08:59:30 [App     ] Closed 0 holes and added 0 new faces
08:59:30 [App     ] Removed 0 non-manifold faces
08:59:30 [App     ] Removed 0 non-manifold vertices
08:59:30 [App     ] Cleaned mesh: 139976 vertices, 278588 faces (650ms)
08:59:30 [App     ] Mesh subdivided: 161331/322135 -> 139976/278588 vertices/faces
08:59:30 [App     ] Mesh saved: 139976 vertices, 278588 faces (21ms)
Processed iterations 45 (100%, 42s418ms)           
09:00:13 [App     ] Mesh saved: 139976 vertices, 278588 faces (22ms)
09:00:13 [App     ] Refine mesh at: 0.50 image scale
09:00:14 [App     ] Mesh subdivided: 139976/278588 -> 160616/319656 vertices/faces
09:00:14 [App     ] Mesh saved: 160616 vertices, 319656 faces (24ms)
Processed iterations 22 (100%, 56s626ms)        ^X^F
09:01:11 [App     ] Mesh saved: 160616 vertices, 319656 faces (28ms)
09:01:11 [App     ] Refine mesh at: 1.00 image scale
09:01:14 [App     ] Mesh subdivided: 160616/319656 -> 331677/660670 vertices/faces
09:01:14 [App     ] Mesh saved: 331677 vertices, 660670 faces (53ms)
Killed

[cdcseacave]

most probably you run out of memory

[sjmduncan]

Hmmmm, memory use sits at ~50% right up until the crash.

[bialasjaroslaw]

The same issue. Machine has 32G of memory available. There are 3 distinguish stages at which process is using ~1/4.7/16.4GB of RAM. Last one is the moment when application crashes before doing or printing anything. System itself is using ~6.5GB of ram sore it is still plenty to use, but somehow it is crashing. I tried to reduce number of steps to two using --scales 2 but without success.

[davidflavigne]

Same issue for me. Memory goes up to ~70%, using https://github.com/rperrot/ReconstructionDataSet.git examples (MuratoCorsicaChurch) processed using openMVG.

$ RefineMesh scene_dense_mesh.mvs --max-face-area 128 --scales 1 -v 5
08:45:04 [App     ] Build date: May  8 2023, 21:30:33
08:45:04 [App     ] CPU: Intel(R) Core(TM) i5-8500 CPU @ 3.00GHz (6 cores)
08:45:04 [App     ] RAM: 7.70GB Physical Memory 16.70GB Virtual Memory
08:45:04 [App     ] OS: Linux 5.15.0-71-generic (x86_64)
08:45:04 [App     ] Disk: 110.03GB (915.82GB) space
08:45:04 [App     ] SSE & AVX compatible CPU & OS detected
08:45:04 [App     ] Command line: RefineMesh scene_dense_mesh.mvs --max-face-area 128 --scales 1 -v 5
08:45:05 [App     ] Scene loaded (157ms):
        42 images (42 calibrated) with a total of 842.19 MPixels (20.05 MPixels/image)
        0 points, 32867 vertices, 65710 faces
08:45:05 [App     ] Refine mesh at: 1.00 image scale
[1]    246492 killed     RefineMesh   128

[cdcseacave]

this is most probably OOM

[davidflavigne]

In my case, RefineMesh was killed when the swap got full. So, indeed a memory problem. Solved by using --resolution-level 1 to downscale processed images and thus limit memory usage
@cdcseacave thanks for answering quickly