This code was created as one of the steps in a research project Dr. Navid Kardan and I worked on over the summer as part of the University of Central Florida's CRCV REU 2022.
Our goal was to recreate the Neural Knitwork architecture found in the paper Neural Knitworks: Patched Neural Implicit Representation Networks. The main difference between our implementation and theirs is that we do not have a multiscale patch representation, we only have one patch scale. We also do not have cross-patch consistency or a discriminator. The code for the patch MLP and Fourier feature mapping was adapted from Jax code for the paper Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains, which can be found here.
Neural Knitworks: Patched Neural Implicit Representation Networks