A working group about Shapes
Comments about the tentative list of the state-of-the-art about shapes
- The original slides are here.
- What for ?
- Consideration of time : development, growth (time evolutions)
- Human perception; psychology; psychophysics
- Geometry (interaction of molecules due to their configuration)
- What ?
- I : What do we mean by shapes ?
- Shapes could be just a list of points, a set of key-points (landmarks)
- The shape might not exist in reality (or the choice of its limits might be arbitrary if several possibilities of definition) but be a practical way of modeling problems
- II : Representation
- A shape can be seen as an infinite number of points (e.g. continuous curve as a set of points)
- Distinction between mathematical models and use in practice (a model could be continuous and need discretization when implemented)
- Related : level-set representation do not imply a discretization as a mathematical model (remove the word "approximation")
- Topology : genus = shape for a particular choice of a wide class of invariance
- Shape as a set of parameters for a particular model
- III: families
- Fractals
- Tubular shapes (as a skeleton + a ball)
- Trees, graphs (e.g. models of skeletons)
- Meshes
- Minimal surfaces
- Shape occultation : parts of a shape, or several shapes overlapping = the same family ?
- Parameterized model
- Pairwise shape comparison
- Similarity vs. distances (do we need distances in the mathematical sense [with triangular inequality, etc.])
- The distance depends obviously on the space of shapes chosen
- For shapes that are 0-level of some function, consider e.g. Bernstein polynomials, distance can be based on how much the graph of the function varies based on the (polynomial) basis coefficients
- Based on eigenvalues of a kernel defined on the shape (the heat kernel typically), as in Bronstein & Bronstein
- About optimization : gradient descent + stochastics
- Where from ?
- Shapes can come from mathematical formulas (e.g. the solution of a problem)
- Shape statistics
- The way of computing statistics depends on the family of shapes studied and their representation
- Parameterized models : statistics on the model parameters