You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
If you find our method/dataset helpful, please consider citing our paper:
@inproceedings{ma2025large,
title={A Large-Scale Dataset of Gaussian Splats and Their Self-Supervised Pretraining},
author={Ma, Qi and Li, Yue and Ren, Bin and Sebe, Nicu and Konukoglu, Ender and Gevers, Theo and Van Gool, Luc and Paudel, Danda Pani},
booktitle={2025 International Conference on 3DV},
pages={145--155},
year={2025},
organization={IEEE}
}
Log in or Sign Up to review the conditions and access this dataset content.
Metadata
As part of the ShapeSplat dataset, the objaverse split contains 141,703 3DGS objects.
Please find the metadata of each 3DGS object, including object_name, psnr, ssim, lpips, num_GS, object_caption, in the completed_3dgs_metadata.csv.
The object captions are from the TRELLIS-500K dataset.
All the 3DGS objects are optimized using rendered images from 72 views, which we released at Objaverse_2d_renders, with 50K gaussians per-obejct.
Each subfolder has the following struture after unzipping:
objaverse_3dgs/000-000/000074a334c541878360457c672b6c2e
├── cfg.yml
├── ckpts
│ └── point_cloud_15000.ply
└── stats
└── val_step15000.json
- Downloads last month
- 18