Closed akenmorris closed 1 year ago
I tried centering, and relative pose variations are lost. The missed correspondence is seen in the global alignment option, but the model looks fine for modes w.r.t domain 1 and 2. We could discuss this use case in our dev meeting.
Running the python script on Linux, all combinations of alignment reference and Multi-level PCA show very poor correspondence for me.
https://user-images.githubusercontent.com/1693349/230780948-62045170-47a1-4a5f-aca1-8d95ebcaea3d.mp4
@iyerkrithika21 , you get different results?
Using "center" alignment in Studio, leaving all other options as is:
https://user-images.githubusercontent.com/1693349/230783121-8708f7c5-dbfd-4388-bd46-0865d4342ebc.mp4
Can you describe more about the relative pose being lost?
@akenmorris: The use case was added before the MLCA additions, so I used the default option for analysis.
The last two gifs here: http://sciinstitute.github.io/ShapeWorks/6.4/use-cases/mesh-based/hip.html shows the relative alignment variations. Is that visible if you center the meshes?
I see. I'm not suggesting that we use 'center' for alignment, I'm just saying that it seems something may be off about how the python script is doing alignment since I get better results using either alignment option in Studio.
Resolved.
The hip multiple domain use case shows very poor correspondence:
https://user-images.githubusercontent.com/1693349/230526326-bffe0576-d297-467d-b2bb-cb287de03fc9.mp4
This is also seen on the documentation page:
http://sciinstitute.github.io/ShapeWorks/6.4/use-cases/mesh-based/hip.html
https://sci.utah.edu/~shapeworks/doc-resources/gifs/hip_global.gif
I think there may be something wrong with the alignment. Using the resulting project, if I go back to grooming in Studio and use either 'center' or ICP and reuse the same optimization parameters, the result better.