HexPlane model implementation based on nerfstudio
As it is not completed yet and requires few (but important) additional implementations, it'll be welcomed if you want to contribute to this repository. (Please refer to "Roadmap" section at bottom of the README)
Please feel free to send an e-mail to me or leave a comment/issue here.
0. Install nerfstudio dependencies and nerfstudio
Refer to the nerfstudio installation document to install nerfstudio dependencies and nerfstudio.
1. Clone this repo
git clone https://github.com/wowo0709/hexplane-nerfstudio.git
2. Install this repo as a python package
Navigate to this folder and run below command.
python -m pip install -e .
3. Run ns-install-cli
This needs to be rerun when the CLI changes, for example if nerfstudio is updated.
ns-install-cli
4. Check the installation
With below command, use should see a list of "subcommands" with hexplane included among them.
ns-train -h
With below command, use should see lists of parameters that we can change while using hexplane.
ns-train hexplane -h
1. Download/Prepare the dataset
For DNeRF dataset, you can use below command. (Refer to nerfstudio document for further details)
It will download the DNeRF dataset under the data folder.
ns-download-data dnerf
If you encounter an error, simply download the dataset here (Thanks to D-NeRF repository)
2. Run hexplane
Run hexplane model with below command. (Refer to nerfstudio document for further details)
ns-train hexplane --data <data_folder>
Qualitative results
Quantitative results
Ablations
Expected future updates to this repository:
- Change beta value of Adam optimizer according to the paper
- Implement TV regularization
- Include other dataset (such as Plenoptic video dataset which was used in the paper)
- Support depth loss