Code NeRF

Code NeRF is a latent variable auto decoder model that learns the variation in shape and texture of objects in a semantic category. Unlike NeRF, Code NeRF disentangles shape and texture by learning separate embeddings. The model is also capable of optimizing the camera viewpoint centered around the object.

Shape and texture embedding optimization

This project is an unofficial implementation of Code NeRF: Disentangled Neural Radiance Field for Object Categories. Since there is no official implementation available, this project may not faithfully reproduce numbers in the paper.

Camera pose optimization

Installation

git clone https://github.com/akashsharma02/code-nerf.git
cd code-nerf/
conda env create -f environment.yml

How to run

1. Download the SRN chairs and cars dataset from the original SRN paper: Google Drive
2. Modify the configuration or create a new configuration using examples from the config/ folder.
3. Run the training script (may take a long time)

python train.py -c config/srn-cars-code.yml

4. Run evaluation script

python eval.py -c config/srn-cars-code-3080-eval.yml

Citation

Please cite the following paper and this implementation if you find this useful:

@InProceedings{Jang_2021_ICCV,
    author    = {Jang, Wonbong and Agapito, Lourdes},
    title     = {CodeNeRF: Disentangled Neural Radiance Fields for Object Categories},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {12949-12958}
}

@misc{sharma2022codenerf,
  title={code-nerf},
  author={Akash, Sharma},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished={\url{https://github.com/akashsharma02/code-nerf.git}},
  month = {March},
  year={2022}
}