Welcome to the svox documentation

This is a PlenOctree volume rendering implementation as a PyTorch extension with CUDA acceleration.

Note this only implements octree (or more generally N^3 tree) operations and differentiable volume rendering. It does not relate to NeRF training part of the project and does not involve a neural network.

The code is available in this repo. This is part of the code release of

PlenOctrees for Real Time Rendering of Neural Radiance Fields

Alex Yu, Ruilong Li, Matthew Tancik, Hao Li, Ren Ng, Angjoo Kanazawa

Paper: https://arxiv.org/abs/2103.14024

Website: https://alexyu.net/plenoctrees

Source code (SVOX): https://github.com/sxyu/svox

Install with pip install svox.

>>> import svox
>>> tree = svox.N3Tree(data_dim=4)
>>> print(tree)
svox.N3Tree(N=2, data_dim=4, depth_limit=10; capacity:1/1 max_depth:0)
>>> tree.to('cuda:0')

data_dim is the size of data stored at each leaf.

Please see Reference for detailed per-method documentation and Quick Guide for a quick overview of features. This is the documentation of svox version: 0.2.32-dev.

Contents:

• Quick Guide
  • Construction
  • Saving and Loading
  • Querying and Modifying Data using N3TreeView
  • Refinement oracle
  • Differentiable Volume Rendering
  • Advanced Leaf-level Accessors
  • Advanced: Volume Rendering Weight Accumulator Context
• Reference
  • N^3-Tree
  • N^3-Tree View
  • Differentiable Volume Renderer
• Example: Lego Bulldozer Rendering
• Example: Optimization and expand