nerf: a reimplementation

A Customized NeRF pipeline by Qi

Install

To install the environment, you just need to type (assume anaconda or miniconda installed),

make setup_env

Or using scripts, after using conda (conda create -n nerf -y python=3.8, conda activate nerf) or it will on your default python environment.

bash ./scripts/env_setting.sh

Or using requirements.txt, python -m pip install -r ./requirements.txt.

Download dataset

To get a minimal dataset for traning demo, you just need to type

make download_data

Or using scripts,

bash ./scripts/download_data.sh

To get more dataset, see data/README.md for details.

Customized dataset

Please refer to preprocess_data/README.md for more information to a customized dataset (currently it was not fully supported).

Trainning

To train a lego for yourself after the environment had been set up, you just need to type

make train_lego

Or using scripts,

bash ./scripts/train_lego.sh

This will make a log directory for check points, visualizations, etc. See makefile for other commands, and see arg_parser.py for more options on trainning.

Download pretrained

You can download pretrained model from our release page, and place then in ./pretrained fold.

Extra your mesh

To extra a mesh yourself, you just need to type

make extra_mesh_demo

Or in extra_mesh.sh

bash ./srcipts/extra_mesh.sh

To be more customized, see extra_mesh/README.md for details.

Please see makefile for more information.

Reference

• NeRF : original project page
• D-NeRF: D-NeRF is based on the same project as this one.
• NeuS