OPONeRF: One-Point-One NeRF for Robust Neural Rendering

TODO List

• Dataset Release
  • Cooking-Perturbation Dataset
  • MeetRoom-Perturbation Dataset
  • Spatial Reconstruction Dataset Links
  • Spatio-temporal Reconstruction Dataset Links
• Code Release
  • Training Code
  • Inference Code

Usage

Setup

pip install -r requirements.txt

Run

bash train_beef.sh

Static Scene Datasets

• Cooking-Perturbation Dataset (Download)

  • Re-organized from N3DV dataset
  • Camera poses estimated using COLMAP

• MeetRoom-Perturbation Dataset (Download)

  • Re-organized from MeetRoom dataset
  • Camera poses estimated using COLMAP

Generalization Datasets

• Generalizable Spatial Reconstruction

  • Available via IBRNet or NeuRay

• Generalizable Spatio-temporal Reconstruction

  • Available via MonoNeRF

The poses of Cooking-Perturbation and MeetRoom-Perturbation were estimated via colmap.

OPONeRF
|-- data
    |--model
        |-- neuray_gen_depth_train_beef_hyper
    |-- example
        |-- beef_0
        |-- beef_1
        ...
        |-- beef_21
    |-- render
        |-- example
            |-- beef_0
            |-- beef_1
            ...
            |-- beef_21
            

Citation

If you find our work useful in your research, please consider citing:

@article{zheng2024oponerf,
  title={OPONeRF: One-Point-One NeRF for Robust Neural Rendering},
  author={Zheng, Yu and Duan, Yueqi and Zheng, Kangfu and Yan, Hongru and Lu, Jiwen and Zhou, Jie},
  journal={arXiv preprint arXiv:2409.20043},
  year={2024}
}