Small Object Photogrammetry via Turntable with Clay Support (2025.1)

[Edward Gonzalez-Tennant]

Hello,

I'm trying to get photo sets that use clay to hold artifacts to properly align. I have photo sets that were taken with the purpose of working in Metashape and/or RealityScan (see image below) that I want to test in Meshroom. 

These sets consist of two groups (chunks) that correspond to the object being oriented in opposite directions. I'm accustomed to the Metahspae/RealityScan workflows where you clip parts of the mesh before aligning and merging chunks (Metashape) or exporting masks from partial models to then use for a final model (RealityScan). 

I've tried several workflows (including some of the new pipelines) but keep running into issues. I'm familiar with the approach by Alban Brice Pimpaud, which makes me think the problem is the clay base in these photo sets? I'm accustomed to using the generate masks features in Metashape/RealityScan to loop through and produce higher quality models. Since that feature/workflow doesn't exist in Meshroom, how do people accomplish this? 

I have also had luck processing each set independently, exporting the models, and completing a full 3D model with Meshlab. I think I'm just missing some steps (and perhaps nodes)? 

Thanks in advance for any help! 

[Fabien Castan]

Hi,

It should work with the pipeline "photogrammetry object two sides" and drag&drop each side of the dataset onto the corresponding CameraInit. To activate a CameraInit in the ImageGallery, double click on the CameraInit node and drag&drop your images.

You also can modify the prompt of the segmentation to be more precise and segment only the "rock" part.

If it does not work, could you describe the issues or share a sample dataset to reproduce the problem?

Best,

--
You received this message because you are subscribed to the Google Groups "AliceVision" group.
To unsubscribe from this group and stop receiving emails from it, send an email to alicevision...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/alicevision/b668365b-2430-4c0a-b170-4876b7df1dd6n%40googlegroups.com.

[Edward Gonzalez-Tennant]

Hello,

I ran the dataset with default settings and it failed at the SfMTriangulation2 node. Here's a Google Drive link to the photoset: https://drive.google.com/file/d/1Rt_hhunGNjosbqhIJ4NKxOwXYQ24TKTE/view?usp=sharing

The masks are still all over the place, some only show the artifact, some show it and the clay, others show it with the clay base and scale bar. I still think the photoset is the issue and that I may need a different approach to the photos for reconstruction using Metashape. Interested to hear what you or others may think. 

I'm processing these on a Windows 11 machine with AMD Ryzen 7 7800X3D that has an onboard AMD Radeon Graphics Process as well as a NVIDIA GeForce RTX 4070 SUPER (GeForce Game Ready Drive installed). I'll try to force GPU use and also run with sift. The defaults include "Use GPU" for ImageSegmentationBox nodes. 

Thanks!

Ed