NeuPhysics: Editable Neural Geometry and Physics from Monocular Videos

Yi-Ling Qiao*, Alexander Gao*, Ming C. Lin

Paper | Project Page | Video

We present a method for learning 3D geometry and physics parameters of a dynamic scene from only a monocular RGB video input.


This code has been tested on Ubuntu 20.04 with CUDA 11.3

Clone this repository and install dependencies. (Recommended to create a new conda env or pipenv.)

git clone
cd NeuPhysics
pip install -r requirements.txt
Dependencies (click to expand)

– torch==1.11.0 – opencv_python== – trimesh==3.9.8 – numpy==1.21.2 – pyhocon==0.3.57 – icecream==2.1.0 – tqdm==4.50.2 – scipy==1.7.0 – PyMCubes==0.1.2 – imageio=2.16.1 – scikit-image=0.19.2


Download data and pretrained model weights on Google Drive.

Extract public_data and exp directories into the top-level neuphysics directory.

Data Convention

The data is organized as follows:

|-- cameras_xxx.npz    # camera parameters
|-- sparse_points_interest.ply    # contains scene ROI
|-- image
    |-- 000.png        # target image for each view
    |-- 001.png
|-- mask
    |-- 000.png        # target mask each view (For unmasked setting, set all pixels as 255)
    |-- 001.png

Here the cameras_xxx.npz follows the data format in IDR, where world_mat_xx denotes the world to image projection matrix, and scale_mat_xx denotes the normalization matrix.


  • Training without mask
python --mode train --conf ./confs/neuphysics_default.conf --case <case_name>
  • Extract surface from trained model for single frame/timestep
python --mode validate_mesh --val_frame_idx <frame_index> --conf <config_file> --case <case_name> --is_continue # use latest checkpoint
  • Extract surface from trained model for entire motion sequence
python --mode validate_mesh_sequence --conf <config_file> --case <case_name> --is_continue # use latest checkpoint

The corresponding mesh can be found in exp/<case_name>/<exp_name>/meshes/<iter_steps>.ply.

  • View interpolation
python --mode interpolate_<img_idx_0>_<img_idx_1> --conf <config_file> --case <case_name> --is_continue # use latest checkpoint

The corresponding image set of view interpolation can be found in exp/<case_name>/<exp_name>/render/.

Train NeuPhysics with your own data

More information can be found in preprocess_custom_data.


If you find this repository helpful, please consider citing our paper as well as others that our work built upon:

    author  = {Qiao, Yi-Ling and Gao, Alexander and Lin, Ming C.},
    title  = {NeuPhysics: Editable Neural Geometry and Physics from Monocular Videos},
    booktitle = {Conference on Neural Information Processing Systems (NeurIPS)},
    year  = {2022},

      title={NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction}, 
      author={Peng Wang and Lingjie Liu and Yuan Liu and Christian Theobalt and Taku Komura and Wenping Wang},

    title = {Non-Rigid Neural Radiance Fields: Reconstruction and Novel View Synthesis of a Dynamic Scene From Monocular Video},
    author = {Tretschk, Edgar and Tewari, Ayush and Golyanik, Vladislav and Zollh\"{o}fer, Michael and Lassner, Christoph and Theobalt, Christian},
    booktitle = {{IEEE} International Conference on Computer Vision ({ICCV})},
    year = {2021},

    author = {Du, Tao and Wu, Kui and Ma, Pingchuan and Wah, Sebastien and Spielberg, Andrew and Rus, Daniela and Matusik, Wojciech},
    title = {DiffPD: Differentiable Projective Dynamics},
    year = {2021},
    journal = {ACM Trans. Graph.},


Our code borrows from NeuS and NonRigid NeRF. Additionally, some code snippets are borrowed from IDR and NeRF-pytorch.

Thank you to the authors of these projects for their great work.


View Github