Meshing fails with No camera to make the reconstruction

[wboerdijk]

Describe the problem I am using meshroom for photogrammetric 3d reconstruction on a set of terrain texture images. The data is not particularly easy, and occasionally some images are not reconstructed (i.e., no camera pose has been found). Yet, the pipeline fails in the Meshing step with the following error:

No camera to make the reconstruction I haven't found any help regarding this issue online, thus I'd appreciate help on this! Is there anything I've forgotten to set up? I've also attached the log below. Thanks in advance for any help!

Log

ERROR:root:Error on node computation: Error on node "Meshing_1":
Log:
Program called with the following parameters:
 * addLandmarksToTheDensePointCloud = 0
 * angleFactor = 15
 * colorizeOutput = 0
 * contributeMarginFactor = 2
 * densifyNbBack = 0 (default)
 * densifyNbFront = 0 (default)
 * densifyScale = 1 (default)
 * depthMapsFolder = "/tmp/MeshroomCache/DepthMapFilter/2a0512c9a29295adb8cb92c2bf41553e0e7feda8"
 * estimateSpaceFromSfM = 1
 * estimateSpaceMinObservationAngle = 10
 * estimateSpaceMinObservations = 3
 * exportDebugTetrahedralization = 0
 * fullWeight = 1
 * helperPointsGridSize = 10
 * input = "/tmp/MeshroomCache/StructureFromMotion/a569d3f33644c53f965b0dc966ce6c57286d4209/sfm.abc"
 * invertTetrahedronBasedOnNeighborsNbIterations = 10
 * maskBorderSize = 1 (default)
 * maskHelperPointsWeight = 0 (default)
 * maxInputPoints = 50000000
 * maxNbConnectedHelperPoints = 50
 * maxPoints = 5000000
 * maxPointsPerVoxel = 1000000
 * minAngleThreshold = 1
 * minSolidAngleRatio = 0.2
 * minStep = 2
 * minVis = 2 (default)
 * nPixelSizeBehind = 4
 * nbSolidAngleFilteringIterations = 2
 * output = "/tmp/MeshroomCache/Meshing/b0161d4ca9fcbf48968233977528dc5ca98cf841/densePointCloud.abc"
 * outputMesh = "/tmp/MeshroomCache/Meshing/b0161d4ca9fcbf48968233977528dc5ca98cf841/mesh.obj"
 * partitioning =  Unknown Type "17EPartitioningMode"
 * pixSizeMarginFinalCoef = 4
 * pixSizeMarginInitCoef = 2
 * refineFuse = 1
 * repartition =  Unknown Type "16ERepartitionMode"
 * saveRawDensePointCloud = 0
 * seed =  Unknown Type "j"
 * simFactor = 15
 * simGaussianSize = 10
 * simGaussianSizeInit = 10
 * universePercentile = 0.999 (default)
 * verboseLevel = "info"
 * voteFilteringForWeaklySupportedSurfaces = 1
 * voteMarginFactor = 4

[16:30:18.099733][info] Found 1 image dimension(s): 
[16:30:18.099828][info]     - [7008x4672]
[16:30:18.104351][info] Overall maximum dimension: [7008x4672]
[16:30:18.104433][warning] repartitionMode: 1
[16:30:18.104443][warning] partitioningMode: 1
[16:30:18.104447][info] Meshing mode: multi-resolution, partitioning: single block.
[16:30:18.104456][info] Estimate space from SfM.
[16:30:18.104831][info] Estimate space done.
[16:30:18.104855][info] bounding Box : length: 3.53292, width: 0.130955, height: 0.131437
[16:30:18.107216][fatal] No camera to make the reconstruction

Desktop (please complete the following and other pertinent information):

• OS: ubuntu 20.02
• Python version 2.7.18
• Qt/PySide version [e.g. 5.12.4]
• Meshroom version: 2021.1.0

[wboerdijk]

any experience with this @fabiencastan @yann-lty @gregoire-dl ? Or is there someone else you could point me to to help sorting this out?

[Fmstrat]

I have just run into this issue using a turntable and CCTAGs if I can be of assistance in debugging.

[Fmstrat]

Additional information: This seems to only occur if I use CCTAGS. Using the same photos with multiple runs, it always fails, however starting a new project without the CCTAG inputs succeeds. Is there any additional info I can provide to help with debugging?

[simogasp]

If you are using CCTAG alone it's likely to fail because there are too few features. For that you need to hack the code and remove some constraints in the sfm part. You can look at this branch https://github.com/alicevision/AliceVision/tree/dev/sfm_markers if you want to see which part (you can use the branch but you may have some updates to make)

[Fmstrat]

@simogasp Thank you, it's not CCTAGs alone, but a model with tags. It's a random starfish thing that I'm using so I can easily measure after to test the CCTAG accuracy. When I don't process the CCTAGs, it works pretty good (edges are rough, likely due to exposure, but otherwise clean). But I get the error when I follow the process for two of the CCTAGs. Attached is a sample shot (217 total). NOTE: Scaled way down and converted RAW to PNG