EvGGS: A Collaborative Learning Framework for Event-based Generalizable Gaussian Splatting

A Collaborative Learning Framework for Event-based Generalizable Gaussian Splatting

Jiaxu Wang, Junhao He, Ziyi Zhang, Mingyuan Sun, Jingkai Sun, Renjing Xu*

Fig 1. The main pipeline overview of the proposed EvGGS framework.

Create environment

conda env create --file environment.yml
conda activate evggs

Then, compile the diff-gaussian-rasterization in 3DGS repository:

git clone https://github.com/graphdeco-inria/gaussian-splatting --recursive
cd gaussian-splatting/
pip install -e submodules/diff-gaussian-rasterization
cd ..

Download models

Download the pretrained models from OneDrive that are placed at \pretrain_ckpt. This directory includes two warmup ckpts and a pretrained ckpts on the synthetic dataset.

Running the code

Download dataset

Ev3D-S

A large-scale synthetic Event-based dataset with varying textures and materials accompanied by well-calibrated frames, depth, and groundtruths.

You can download the dataset from OneDrive and unzip it. A 50 GB of storage space is necessary.
EV3D-R

A large-scale realistic Event-based 3D dataset containing various objects captured by a real event camera DVXplore.

You can download the dataset from OneDrive and Baidu Disk, code:3ilt.

Training

python train_gs.py

Evaluation

python eval_gs.py

In configs\Ev3D_pretrain, several primary settings are defined such as experimental name, please check.

Citation

please cite our work if you use this dataset.

@misc{wang2024evggscollaborativelearningframework,
      title={EvGGS: A Collaborative Learning Framework for Event-based Generalizable Gaussian Splatting}, 
      author={Jiaxu Wang and Junhao He and Ziyi Zhang and Mingyuan Sun and Jingkai Sun and Renjing Xu},
      year={2024},
      eprint={2405.14959},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2405.14959}, 
}

Reference

EventNeRF: https://github.com/r00tman/EventNeRF?tab=readme-ov-file. 3D Gaussian Splatting: https://github.com/graphdeco-inria/gaussian-splatting. GPS-GS: https://github.com/aipixel/GPS-Gaussian PAEvD3d: https://github.com/Mercerai/PAEv3d

Mercerai / EvGGS