RHINO: Regularizing the Hash-based Implicit Neural Representation

PyTorch implementation of RHINO.

Pipeline

Setup

We provide a conda environment setup file including all of the above dependencies. Create the conda environment RHINO by running:

conda env create -f Rhino.yaml

Task

Image Representation

• Training

For tasks like fitting a single image, we prepare a test image in the data folder.

To train image representations, use the config files in the config folder. For example, to train on the provided image, run the following

cd img
python train_img_interp.py --config ./config/img.ini

After the image representation has been trained, the results of the image will appear in the img/log/<experiment_name> folder, where <experiment_name> is the subdirectory in the log folder corresponding to the particular training run.

3D Shape Representation

• Datasets

  Datasets can be downloaded using the sdf/download_datasets.py script.

• Training

cd sdf/experiments
python train_sdf_RHINO.py --config ./config/sdf/bacon_armadillo.ini

After the 3D shape representation is trained, the results will appear in the sdf/log/<experiment_name> folder

• Rendering

cd sdf/experiments
python render_sdf_RHINO.py

You can obtain the rendering results of different data by modifying bacon_names.

Citation

@article{zhu2023rhino,
      title={RHINO: Regularizing the Hash-based Implicit Neural Representation},
      author={Zhu, Hao and Liu, Fengyi and Zhang, Qi and Cao, Xun and Ma, Zhan},
      journal={arXiv preprint arXiv:2309.12642},
      year={2023}
    }