ULB SauceDino

ULB SauceDino synthetic light field dataset by LISA ULB
The synthetic light field dataset "ULB SauceDino" [1] is provided by Armand Losfeld, Laurie Van Bogaert, Gauthier Lafruit, Mehrdad Teratani, members of the LISA department, EPB (Ecole Polytechnique de Bruxelles), ULB (Université Libre de Bruxelles), Belgium.
Content
This dataset is used for the performance evaluation of tensor displays, but any other applications using a light field (or multi-views) as input can use it. It is composed of 3 scanned objects from the collection Scanned Objects by Google Research [2], two user-modeled objects, and a brick-texture background. Two pre-rendered light fields are provided. If other light field configurations are needed, please install Blender ($\geq$ 3.8.0) and the light field blender add-on (link: https://github.com/dbonattoj/blender-addon).
Please find a detailed description of the content of each file in the following sections.
9x9_wide_FOV.zip
This compressed file contains 9 horizontal by 9 vertical viewpoints for a total of 81 viewpoints. Each viewpoint contains 512x512 pixels and is stored in the PNG format with 8 bits for each channel. The camera parameters, cf. parameters.cfg, are set to have a disparity of -3 in the background, a disparity of 0 in the dinosaur body, and a disparity of 3 in the letter cube. This light field is generally used to challenge the evaluated method due to its high-parallax and high-texture objects.
Note that, the viewpoint called Cam000 is always the top-left viewpoint of the light field. The next viewpoint Cam001 is the adjacent viewpoint on the right. So, the (N-1)-th viewpoint (here 80) is the bottom-right viewpoint.
15x15.zip
This compressed file contains 15 horizontal by 15 vertical viewpoints for a total of 125 viewpoints. Each viewpoint contains 512x300 pixels and is stored in the PNG format with 8 bits for each channel. The camera parameters, cf. parameters.cfg, are set to have a disparity of -1 in the background, a disparity of 0 in the dinosaur body, and a disparity of 1 in the letter cube. Due to its low parallax, this light field is easier to reproduce than the previous one.
Note that, the viewpoint called Cam000 is always the top-left viewpoint of the light field. The next viewpoint Cam001 is the adjacent viewpoint on the right. So, the (N-1)-th viewpoint (here 224) is the bottom-right viewpoint.
BLENDER_SCENE.zip
This compressed file contains all the files needed to load the scene in Blender. With this, it is possible to render any new light fields of the same scene and even change the complexity of the scene. Note that, the cameras used for the previous light field rendering are already present. We only recommend the modification of the camera parameters instead of their position since only the adjustable camera parameters can be exported with the Blender add-on in the parameters.cfg file.
License
CC BY-NC-SA
Terms of Use
Any kind of publication or report using this dataset should refer to the references below.
References
[1] Armand Losfeld, Laurie Van Bogaert, Gauthier Lafruit, Mehrdad Teratani,"ULB SauceDino", 2023.
@misc{losfeld_saucedino_2023, title = {{ULB} {SauceDino}}, author = {Losfeld, Armand and Van Bogaert, Laurie and Lafruit, Gauthier and Teratani, Mehrdad}, month = may, year = {2023}, doi = {10.5281/zenodo.7950729} }
[2] L. Downs, A. Francis, N. Koenig, B. Kinman, R. Hickman, K. Reymann, T. B. McHugh, and V. Vanhoucke, "Google Scanned Objects: A High-Quality Dataset of 3D Scanned Household Items," 2022. Available: https://arxiv.org/abs/2204.11918
@misc{downs2022google, title={Google Scanned Objects: A High-Quality Dataset of 3D Scanned Household Items}, author={Laura Downs and Anthony Francis and Nate Koenig and Brandon Kinman and Ryan Hickman and Krista Reymann and Thomas B. McHugh and Vincent Vanhoucke}, year={2022}, eprint={2204.11918}, archivePrefix={arXiv}, primaryClass={cs.RO} }