Link Search Menu Expand Document (external link)

NeRF-Factory

An awesome NeRF collection.

Installation View it on GitHub


logo

Motivation

Neural fields has recently been one of the most discussed topics in computer vision. As seen by a large number of NeRF papers, we can realize that it has been developed rapidly. However, we have not found a large-scale library that collects multiple NeRF implementations into an integrated form. Thus, this library, called NeRF-Factory, provides a convenient tool for evaluating and comparing NeRF models. Our library is super-easy to add your custom data and model by integrating format of codes.

Supports

  • Allows multi-GPU training with PyTorch-Lightning except for models that are even inefficient when running with Multi-GPU, such as Plenoxels and DVGO.
  • A project page that includes instructions, results, and links for the pretrained models.
  • Dividing the NeRF’s training process into three phases: ray generation, network forwarding, and optimization.
  • Convenient switching of respective process to desired options by simply switching the config.
  • Interactive visualization of rendered images for convenient comparison between trained models.