NeRF-Tex

This repo contains the implementation of NeRF-Tex: Neural Reflectance Field Textures written in Tensorflow. The paper and EGSR presentation can be found on the project page. Should you have a question regarding the implementation feel free to contact Hendrik Baatz.

Setting up the code for use

You'll need to have Conda and Embree 3 installed on your system. Then just follow the steps below with the working directory being the one where this README file resides.

Create & activate the python environment

conda env create
conda activate nerf-tex

Get the dependencies in case you did not already clone this repo with the --recursive flag
```
git submodule update --init
```
Build the instancer code on Linux (you may want to adapt the Embree include and library dirs in instancer/setup.py depending on your local setup)
```
make -C instancer
```

Creating the example dataset

We provide an example showing how our datasets are created. You'll need to have Blender (2.92) installed and in your path. This takes a while.

sh data/create_dataset.sh data/carpet.blend data/configs/config_carpet.py
python data/nerf2tfr.py datasets/materials/carpet dataset/materials/carpet/tfr

Downloading the datasets

The datasets are available here.

Training an example scene

python main.py configs/config_carpet_train.py

Rendering an example scene

python main.py configs/config_carpet_render.py

Acknowledgements

We thank the contributors of the following projects used in this implentation for making their work available publicly: