BARD-GS: Blur-Aware Reconstruction of Dynamic Scenes via Gaussian Splatting

This as an official implementation of our CVPR 2025 paper BARD-GS: Blur-Aware Reconstruction of Dynamic Scenes via Gaussian Splatting.

Quickstart

1. Installation

# create a conda environment
conda create -n BARD-GS -y python=3.10
conda activate BARD-GS

# install dependencies
pip install --upgrade pip setuptools
pip install torch==2.1.2 torchvision==0.16.2 --index-url https://download.pytorch.org/whl/cu118


conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit
pip install ninja git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

# install nerfstudio
pip install nerfstudio==1.0.3

Then you can clone this repo and install:

git clone https://github.com/luyr/BARD-GS.git
cd BARD-GS
pip install -e .

2. Prepare the dataset

BARD-GS Real World Dataset

Download processed dataset from this link.

Synthetic Dycheck Motion-Blur Dataset

To be released

3. Training

1. For BARD-GS Real World dataset, train with:

   ns-train BARD-GS \
    --data <path to dataset>/card/ \
    --experiment_name <your exp name> \
    --vis wandb bard-gs-data;

2. For Dycheck synthetic motion-blurdataset :

   ns-train BARD-GS \
    --data <path to dataset>/paper-windmill/ \
    --experiment_name <your exp name> \
    --vis wandb deblur-gs-data;

4. Test Set Rendering

ns-render dataset \
--load-config <your training config>  \
--split test \
--rendered-output-names rgb \
--output-path ./renders/<your exp name> \
--image-format png

5. Debug with your IDE

Open this repo with your IDE, create a configuration, and set the executing python script path to <nerfstudio_path>/nerfstudio/scripts/train.py, with the parameters above.

6. TODO

• BARD-GS real world dataset
• Training script
• Dycheck synthetic dataset
• Evaluation script
• Preprocessing script

Citation

If you find this useful, please consider citing:

@inproceedings{lu2025bard,
  title={Bard-gs: Blur-aware reconstruction of dynamic scenes via gaussian splatting},
  author={Lu, Yiren and Zhou, Yunlai and Liu, Disheng and Liang, Tuo and Yin, Yu},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={16532--16542},
  year={2025}
}

Acknowledgments