File size: 1,934 Bytes
730ad3d
 
 
 
 
 
 
5b82f88
730ad3d
5b82f88
730ad3d
 
 
2cd87d6
730ad3d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
---
license: cc-by-sa-4.0
tags:
- olat
pretty_name: 'Dataset for BiGS: Bidirectional Primitives for Relightable 3D Gaussian Splatting'
---

# BiGS Dataset

The OLAT dataset used in the paper _BiGS: Bidirectional Primitives for Relightable 3D Gaussian Splatting_ (3DV 2025). Check out our [project page](https://desmondlzy.me/publications/bigs/).

We provide 7 synthetic scenes in the dataset, featuring various complex light transport effects, such as subsurface scattering, fuzzy surfaces, and iridescent reflection. 

Each scene (1.8 ~ 3.2 GB) consists of:
- 40 training OLAT conditions (`olat_1` - `olat_40`) with 48 views per light condition;
- 58 test OLAT conditions (`olat_41` - `olat_98`) with 1 view per light condition;
- 1 all-light-on conditions (`olat_all`) with 48 views per light conditions.

Each light condition includes `.exr` images, object masks, transforms with camera poses, light positions and intensities. 

Please refer to our [github repo](https://github.com/desmondlzy/bigs) for how to use the dataset provided here to train BiGS, 
and our [paper (arxiv)](https://www.arxiv.org/abs/2408.13370) for details of BiGS.

## Citation

If you use our dataset in your research, please consider citing us with the below bibtex entry:
```
@misc{zhenyuan2024bigs,
      title={BiGS: Bidirectional Primitives for Relightable 3D Gaussian Splatting}, 
      author={Liu Zhenyuan and Yu Guo and Xinyuan Li and Bernd Bickel and Ran Zhang},
      year={2024},
      eprint={2408.13370},
      url={https://arxiv.org/abs/2408.13370}, 
}
```

## Acknowledgments

Our synthetic data is generated using [Mitsuba](https://mitsuba.readthedocs.io/en/stable/).
We thank the 3D models' creators: 
Keenan Crane for _Spot_;
Stanford Computer Graphics Laboratory for the models _Dragon_ and _Bunny_; 
Wenzel Jakob for the model _Mistuba Ball_.
Special thanks to Changxi Zheng for supporting the internship program at Tencent Pixel Lab.