A linear meta-layer that can deal with batched weight matrices and biases, as for instance output by a hypernetwork.

A fully connected neural network that also allows swapping out the weights when used with a hypernetwork. Can be used just as a normal neural netw

Module to add positional encoding as in NeRF [Mildenhall et al. 2020].

A helper class for managing registering modules, it extends a dictionary and provides a register functions. Eg. creeting a registry:

A canonical representation network for a BVP.