This is official code of CVPR2025 paper RoGSplat: Learning Robust Generalizable Human Gaussian Splatting from Sparse Multi-View Images.
sh ./install.sh
Please follow sherf to download the RenderPeople dataset. Unzip and rename it into renderpeople, split dataset into train and val.
Download our estimated and preprocessed data from here and put it under renderpeople folder.
The final "renderpeople" folder should be like this:
renderpeople
├-- train
├-- val
├-- easymocap_smpl
Make a soft link:
ln -s /path/to/renderpeople ./renderpeople
Train first stage:
python train_stage1.py
Then change stage1_ckpt to ./experiments/rp_xxxx/ckpt/rp_final.pth, change depth_ckpt to ./experiments/rp_xxxx/ckpt/rp_depth_latest.pth.
Then train second stage:
python train_stage2.py
python test_stage2.py --ckpt_path experiments/rp_xxxx
If you find this code useful for your research, please cite this:
@inproceedings{RoGSplat2025CVPR,
title={{RoGSplat}: Learning Robust Generalizable Human Gaussian Splatting from Sparse Multi-View Images},
author={Xiao, Junjin and Zhang, Qing and Nie, Yongwei and Zhu, Lei and Zheng, Wei-Shi},
booktitle={CVPR},
year={2025}
}