Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    HEMONCedric
    @HEMONCedric
    Hello,
    I'm a student in my 5th year in France and I'm working on a medical segmentation project and we want to use an SSM to introduce a priori information.
    I'm using the scalismo library to build my SSM model but I have a problem when applying the GPA because my different models have a different number of points. I wanted to know if it was possible to constrain the number of points when i use "val meshFiles = newjava.io.File("D:/NIH_Pancreas/Pancreas_label_stl/").listFiles.take(5) meshFiles.foreach(file => println(s"\n$file"))"
    HEMONCedric
    @HEMONCedric
    val (meshes) = meshFiles.map(meshFile => { val mesh = MeshIO.readMesh(meshFile).get
    (mesh) }) * sorry for the mistake
    Marcel Luethi
    @marcelluethi
    Hi Cedric, You cannot build an ssm with meshes that are not in correspondence (i.e. have the same number of points and the points are at the (semantically ) same location. You will need to do a registration as a first step. How this can be done in Scalismo is explained in this tutorial
    HEMONCedric
    @HEMONCedric
    Okay thank you very much, I will try to use the ICP to solve this problem.
    AngelaCsz
    @AngelaCsz
    Hi,
    image.png
    AngelaCsz
    @AngelaCsz
    If I use this instead of the Gaussian process in the 12th tutorial, can I then replace all the lowRankGP variables of the tutorial by my gpInterpolator ?
    AngelaCsz
    @AngelaCsz
    Also, can I use the registration for complex shapes to create a fitting (and not a projection) ?
    Marcel Luethi
    @marcelluethi
    When you load a shape model from disk, it constructs a DiscreteLowRankGaussianProcess internally (which you can access using pdm.gp). Interpolating the DiscreteLowRankGaussianProcess, as done in the code snippet above, yields a LowRankGaussianProcess. This can be used whereever a LowRankGaussianProcess is accepted as a parameter.
    To your second question: Yes, if you just don't do the projection step, you end up with the fitting result. This works for arbitrarily complex shape, given that you can successfully represent the shape variability with your model
    AngelaCsz
    @AngelaCsz
    Thank you very much !