Geometry primitives

[KonstantinUshenin]

We need geometrical primitives in math_core.py. The code should provide implicit representation of shapes.

Examples:

• https://github.com/ml-for-gp/jaxgptoolbox/blob/de15f5c27af44b6a11b1e087e602d57ce03c378d/general/sdf_circle.py
• https://github.com/ml-for-gp/jaxgptoolbox/blob/de15f5c27af44b6a11b1e087e602d57ce03c378d/general/sdf_triangle.py

List of expected 3D objects:

• Sphere
• Box
• Tetrahedron (equilateral)

List of expected 2D objects:

• Circle
• Square
• Triangle (equilateral)

[KonstantinUshenin]

@hanuel-89 , could you look at this issue? I think it is an easy task for one more contribute.

[hanuel-89]

Sure. I'll look into it.

[KonstantinUshenin]

@hanuel-89 I want to tell you more about the maintainer goals. We want to implement as many as possible methods from computational anatomy using neural networks. Computational anatomy includes a lot of heterogeneous tasks. There are some examples: https://team.inria.fr/epione/en/research/computational-anatomy/ https://en.wikipedia.org/wiki/Computational_anatomy

We are forced to use JAX+HAIKU because PyTorch and Tensorflow are really bad choices for physically informed neural networks and implicit representation.

3D objects may be presented for calculation as voxel models, surface mesh, or point clouds. Implicit representation with neural networks is an advanced approach. The key concept for our code is the signed distance function.

[KonstantinUshenin]

@hanuel-89 , I have just added sdf_primitive_sphere here https://github.com/KonstantinUshenin/nndt/blob/main/nndt/math_core.py
It is the first primitive in this list.

[KonstantinUshenin]

Separate this to #128 #127 #126