overfitSDF

Implementation of the architecture presented in "Overfit Neural Networks as a Compact Shape Representation".

Install

Our main two dependencies are on tensorflow for neural network implementation and libigl for all geometry processing related tasks.

We package all of our dependencies in the included environment.yaml file which can be installed using conda.

conda env create -f environment.yml

Once created the environment can be activated with

conda activate overfitSDF

From now on we assume all commands are ran from within this conda environment.

Usage

Convert Mesh to Weight-encoded neural implicit

python train.py data/bumpy-cube.obj

NOTE: cube marching is not currently implemented, this repo is a work in progress as we port code from libigl C++ to libigl-python-bindings for easier usage. We apologize for the delay!