Correspondence among Synthetic Data Instances

Hello everyone,

I was wondering whether there is a way to obtain particle correspondence for data, such as the one depicted in the attached file?

Any recommendation is much appreciated!


There are multiple ways to interpret this data. In the most basic interpretation, there is no variation here at all, only a change in position (e.g. rotation). Potentially an ICP registration would pre-align the shapes and the shape model would yield nothing much.

Assuming you don’t perform any alignment and disable procrustes alignment as well, you might see results similar to the “box bump” example. For example:

Indeed, I did not perform any ICP alignment or Procrustes alignment, unfortunately, I could not find a parameter combination that result in particle correspondence :frowning_face:

Do you have/know a parameter recommendation that I can take as a starting point?

Looking more closely at your data and discussing with the ShapeWorks team, it seems that the variation in your data is more of a non-linear case. Currently, this can’t be handled by ShapeWorks as it assumes linear trajectory (or linear/gaussian in the shape space). We are currently working on a non-linear extension, but it is not ready yet for a release.

@hannesk95, if possible, could you share this torus bump dataset or a script that was used to generate it?

@iyerkrithika, I’ve uploaded the torus bump data, please use the following link to download the data.

1 Like

Thanks for the data!