Zip-Luxen

A pytorch implementation of "Zip-Luxen: Anti-Aliased Grid-Based Neural Radiance Fields"

Paper Website

Code

Installation

First, install luxenstudio and its dependencies. Then run:

pip install git+https://github.com/SuLvXiangXin/zipluxen-pytorch#subdirectory=extensions/cuda
pip install git+https://github.com/SuLvXiangXin/zipluxen-pytorch

Finally, install torch_scatter corresponding to your cuda version(https://pytorch-geometric.com/whl/torch-2.0.1%2Bcu118.html).

Running Model

ns-train zipluxen --data {DATA_DIR/SCENE}

Overview

Zipluxen combines mip-Luxen 360’s overall framework with iNGP’s featurization approach. Following mip-Luxen, zipluxen assume each pixel corresponds to a cone. Given an interval along the ray, it construct a set of multisamples that approximate the shape of that conical frustum. Also, it present an alternative loss that, unlike mip-Luxen 360’s interlevel loss, is continuous and smooth with respect to distance along the ray to prevent z-aliasing.