Skip to content

HannahHaensen/DynaMoN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DynaMoN: Motion-Aware Fast and Robust Camera Localization for Dynamic Neural Radiance Fields

This is the official code for our paper DynaMoN: Motion-Aware Fast and Robust Camera Localization for Dynamic Neural Radiance Fields.

The accurate reconstruction of dynamic scenes with neural radiance fields is significantly dependent on the estimation of camera poses. Widely used structure-from-motion pipelines encounter difficulties in accurately tracking the camera trajectory when faced with separate dynamics of the scene content and the camera movement. To address this challenge, we propose Dynamic Motion-Aware Fast and Robust Camera Localization for Dynamic Neural Radiance Fields (DynaMoN). DynaMoN utilizes semantic segmentation and generic motion masks to handle dynamic content for initial camera pose estimation and statics-focused ray sampling for fast and accurate novel-view synthesis. Our novel iterative learning scheme switches between training the NeRF and updating the pose parameters for an improved reconstruction and trajectory estimation quality. The proposed pipeline shows significant acceleration of the training process. We extensively evaluate our approach on two real-world dynamic datasets, the TUM RGB-D dataset and the BONN RGB-D Dynamic dataset. DynaMoN improves over the state-of-the-artboth in terms of reconstruction quality and trajectory accuracy.

Architecture of DynaMoN

Installation

Clone the repository

git clone https://github.com/HannahHaensen/DynaMoN.git

Creating a new anaconda environment and installing pytorch and necessary libraries

Create conda environment:

conda create -n dynamon python=3.8
conda activate dynamon

Then install pytorch:

pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 --extra-index-url https://download.pytorch.org/whl/cu117
pip install "git+https://github.com/facebookresearch/pytorch3d.git@stable"

and afterwards the necessary libraries:

pip install -r requirements.txt
pip install evo --upgrade --no-binary evo

Compile the extensions (can take up to several minutes)

conda install cuda-nvcc=11.7 -c nvidia
python setup.py install

Download the model from google drive (droid.pth) and put it into the project root folder.

Compile DCNv2

cd src/DytanVO/Network/rigidmask/networks/DCNv2
python setup.py install

Unzip this folder in the project root folder. There should now be a folder "models".

Evaluation

We provide evaluation scripts for TUM RGB-D and Bonn RGB-D Dynamic datasets.

TUM RGB-D (exemplary for freiburg3_sitting_xyz):

python evaluation_scripts/train_tum.py --datapath /path/to/data/rgbd_dataset_freiburg3_sitting_xyz --disable_vis --filter_motion --filter_semantics

Bonn RGB-D (exemplary for rgbd_bonn_crowd):

python evaluation_scripts/train_bonn.py --datapath /path/to/data/rgbd_bonn_crowd --disable_vis --filter_motion --filter_semantics

Acknowledgement

Our work is based on these different codebases that are thankfully available open-source: HexPlane, DROID-SLAM and DytanVO. We also use the relative rotation accuracy metric calculation from PoseDiffusion. Please respect their licenses when using the code.

If you use our code or paper for your work, please consider citing:

@ARTICLE{10777295,
  title={DynaMoN: Motion-Aware Fast and Robust Camera Localization for Dynamic Neural Radiance Fields}, 
  author={Schischka, Nicolas and Schieber, Hannah and Karaoglu, Mert Asim and Gorgulu, Melih and Grötzner, Florian and Ladikos, Alexander and Navab, Nassir and Roth, Daniel and Busam, Benjamin},
  journal={IEEE Robotics and Automation Letters}, 
  year={2024},
  volume={},
  number={},
  pages={1-8},
  keywords={Cameras;Dynamics;Neural radiance field;Location awareness;Simultaneous localization and mapping;Training;Robot vision systems;Accuracy;Semantics;Optimization;Localization;Mapping},
  doi={10.1109/LRA.2024.3511399}}

License

This codebase is licensed under Creative Commons Attribution-NonCommercial 4.0 International

About

Motion-Aware Fast And Robust Camera Localization for Dynamic NeRF

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published