Still training a neural field with a regression loss and hoping for it to overfit? What if we enforce the regression loss to be zero as a hard constraint in the formulation of a neural field?

This repository is the official implementation of the paper Neural Fields with Hard Constraints of Arbitrary Differential Order, NeurIPS 2023.

• Python 3.x

pip install -r requirements.txt

python fit_Fermat.py

First, download the MERL dataset from https://www.merl.com/brdf/, replace PATH/TO/MERL/DATASET at line 203 of fit_brdf.py with the path to the MERL dataset, then run

python fit_brdf.py

python fit_exact_normal.py

python fit_pointcloud_3d.py

python fit_advection.py