LoopGaussian: Creating 3D Cinemagraph with Multi-view Images via Eulerian Motion Field

Jiyang Li1*, Lechao Cheng2*, Zhangye Wang1, Tingting Mu3, Jingxuan He2✉
*Equal contribution. Corresponding author.
1Zhejiang University, 2Hefei University of Technology, 3The University of Manchester

Environment

  • CUDA = 12.1
  • Pytorch = 2.1.1
  • Python = 3.11
pip install -r requirements.txt

Train

Train 3D-GS

python gaussian_splatting/train.py -s {images_path}

Train LoopGaussian

python train_loop.py --task_name ${task_name}

The rendering result will be saved in output/{task_name}/visual.mp4