MVPGS: Excavating Multi-view Priors for Gaussian Splatting from Sparse Input Views

ECCV 2024 (arXiv | project page | code)

This repository holds the results of MVPGS (optimized models and rendered images), and the MVSFormer weights used in our DTU experiments: MVSFormer trained without the DTU scenes that overlap our selected test scenes, then fine-tuned on BlendedMVS.

Files

File	Size
results/DTU/DTU3_2x.zip	1.66 GiB
results/DTU/DTU3_4x.zip	0.94 GiB
results/LLFF/LLFF3_4x.zip	2.35 GiB
results/LLFF/LLFF3_8x.zip	2.58 GiB
results/NVS-RGBD/kinect3_2x.zip	1.27 GiB
results/NVS-RGBD/zed3_2x.zip	1.68 GiB
results/TanksandTemples/Tanks3_1x.zip	6.42 GiB
MVSFormerWeights/MVSFormer-dtuless-blendedmvs.zip	0.57 GiB

Download

Click a file under "Files and versions", or use the hf CLI:

# everything
hf download zezeaaa/MVPGS --repo-type dataset --local-dir MVPGS
# one file
hf download zezeaaa/MVPGS results/LLFF/LLFF3_8x.zip --repo-type dataset --local-dir MVPGS

Citation

@inproceedings{xu2024mvpgs,
  title={Mvpgs: Excavating multi-view priors for gaussian splatting from sparse input views},
  author={Xu, Wangze and Gao, Huachen and Shen, Shihe and Peng, Rui and Jiao, Jianbo and Wang, Ronggang},
  booktitle={European Conference on Computer Vision},
  pages={203--220},
  year={2024},
  organization={Springer}
}