This is an implementation of a dynamic nerf that adds extra constraints to hopefully improve the performance of dynamic nerfs. This project uses nerfacc for rendering the nerf and its implementation of DNeRF as a base to work on.

This research project was unsuccessful in creating a better-performing NeRF before the dissertation deadline. As I no longer have access to large enough GPUs to run this model, I cannot develop it further. If you'd like to take this further, please read the results section of the dissertation in the release section for the current state of the code.

This project uses the nerfacc module for rendering the NeRF and example code from the repository, which is found here: https://github.com/KAIR-BAIR/nerfacc.

To run the code, you must have an environment setup with Cuda enabled, PyTorch, Torchdiffeq and Nerfacc. As well as this, the dnerf-synthetic dataset will need to be obtained and can be found here https://www.dropbox.com/s/0bf6fl0ye2vz3vr/data.zip?dl=0. This will then need to be extracted to "/home/ruilongli/data/dnerf/". Once the environment is set up you can run python3 main.py followed by arguments to start training/rendering.