This repository is presented as a demonstration program for the Federated NeRF paper arXiv:2305.01163.

It is forked from the krrish94/nerf-pytorch repository.

The key contribution is the impelementation of Algorithm 2 from the paper in train_nerf_federated.py The update compression implementation can be found in update_compression.py.

The lego dataset is included as a demonstration. You can run the demonstration program as follows:

$ pip install -r requirements.txt
$ python train_nerf_federated.py --config demonstration/lego.yml 

The trained networks can then be evaluated, generating CSV files of validation metrics and rendered images, by running

$ ./demonstration/process_results.sh ./demonstration_results/lego 19999 04999