The neural implicit lab is a helper library built on top of pytorch for training implicit neural representations (INR) of geometrical objects. The goal of the library is to make the computation of an INR the ressemble a config file as much as possible and abstract away the data sampling, the training loop and the model definition, while remaining highly customizable.
The library is under construction. Expect heavy changes.
The documentation can be found here: https://GCoiffier.github.io/neural_implicit_lab/
- All input geometries are centered around zero and scaled down so that they fit inside
[-1, 1]^n - By default, datasets of points are sampler inside the
[-1.5, 1.5]^ndomain
- Rely on
meshiofor faster geometry loading - Implement SAL and SALD
- Setup the query module
- Find a better name for the lib :)