MoNeRF: Fast Non-Rigid Radiance Fields from Monocularized Data

Official PyTorch Implementation of 'Fast Non-Rigid Radiance Fields from Monocularized Data'

Getting Started

• Clone this repository using:

git clone https://github.com/MoritzKappel/MoNeRF.git && cd MoNeRF

• Before running our code, you need to install all dependencies listed in scripts/createCondaEnv.sh under # dependencies, or create a new conda environment by executing the script:

./scripts/createCondaEnv.sh && conda activate monerf

• To install the necessary custom CUDA kernels, run:

./scripts/install.py -e VolumeRenderingV2

Creating a Configuration File

Default configurations for the D-NeRF and MMVA datasets are available in the configs/ directory. To create a custom configuration file, run

./scripts/createDefaultConfig.py -m MoNeRF -d MMVA -o <my_config>

and edit the values in configs/<my_config>.yaml as needed.

Dataset

You can manually download our MMVA dataset from here, or run

./scripts/downloadDataset.sh

to automatically download and unpack the sequences to the MoNeRF/dataset directory.

Training a New Model

To train a new model from a configuration file, run:

./scripts/train.py -c configs/<my_config>.yaml

The resulting images and model checkpoints will be saved to the output directory.

To train multiple models from a directory or list of configuration files, use the scripts/sequentialTrain.py script with the -d or -c flag respectively.