ULB Voronoi - Simulated Plenoptic 2.0

The sequence "ULB Voronoi - Simulated Plenoptic 2.0" is provided by Daniele Bonatto, Sarah Fachada, Gauthier Lafruit, members of the LISA department, EPB (Ecole Polytechnique de Bruxelles), ULB (Universite Libre de Bruxelles), Belgium. Licence: CC BY-NC-SA Terms of Use Any kind of publication or report using this sequence should refer to the following references. [1] Daniele Bonatto, Sarah Fachada, Gauthier Lafruit, "ULB Voronoi - Simulated Plenoptic 2.0", 2024. @misc{bonatto_voronoi_2024, title = {{ULB} {Voronoi} - {Simulated} {Plenoptic} {2.0}}, author = {Bonatto, Daniele and Fachada, Sarah and Lafruit, Gauthier}, month = feb, year = {2024}, doi = {10.5281/zenodo.10679358} } [2] D. Bonatto, « From multi-modal capture to photo-realistic view synthesis - A high-quality and real-time multiview approach », Université Libre de Bruxelles, 2024. @thesis{bonatto_multi-modal_2024, location = {Brussels, Belgium}, title = {From multi-modal capture to photo-realistic view synthesis}, institution = {Universite Libre de Bruxelles}, type = {phdthesis}, author = {Bonatto, Daniele}, date = {2024-03}, } Production Laboratory of Image Synthesis and Analysis, LISA department, Ecole Polytechnique de Bruxelles, Universite Libre de Bruxelles, Belgium. Content This dataset contains a static scene (Voronoi) created using the light field blender plugin [4]. We provide calibration images of a squared checkerboard and white images that could be obtained with different main lens apertures. We provide the blender files for the Voronoi dataset. The dataset we generated follows the principle that each micro-lens functions like a pinhole camera. To achieve this, we utilized Blender's light-field plugin [4] to generate an array of cameras that captured a basic scene, thereby simulating the plenoptic camera acquisition process. The Voronoi scene consists of two planes that possess a Voronoi texture and are positioned at distances of $|t_1|=2$mm and $|t_2|=2.5$mm from the \tip{mla}. In order to acquire maximum information, the plenoptic images must be arranged in a typical hexagonal layout [5]. However, the utilized plug-in only allows for rectangular shapes and necessitates equal distances between cameras both horizontally and vertically. Conversely, in hexagonal grids, each row comprises contiguous images, and the vertical spacing has a factor of $\sqrt3/2$ to the horizontal spacing. To account for this discrepancy, we created a $1\times31$ row of contiguous cameras, spaced $B=0.2$mm apart, and shifted it vertically by the corresponding step during 20 frames. To generate the rows between the micro-lenses in the hexagonal arrangement, we rendered a second light-field of the same scene with a shift of $(0.5,\sqrt3/2)B$, resulting in a total of $20\times31$ views per light-field. We established the camera parameters as follows: the micro-lenses were positioned in parallel and possessed a focal length of $f=100$mm, a sensor size of $\text{sensor}=35$mm, a diameter of $D=30$pix, and a baseline of $B=0.2$mm. These specifications correspond to a value of $s=\frac{Bf}{\text{sensor}}=0.5714$mm in a plenoptic camera. We developed a script that uses the two light-fields to produce a hexagonal grid and applies a circular mask to generate an image that mimcs the plenoptic ones. The orientation of the micro-image determines whether the resulting plenoptic camera is in Galilean or Keplerian configuration, corresponding to $t_1=\pm2$mm and $t_2=\pm2.5$mm. The obtained Galilean plenoptic image has dimensions of $1095\times945$. The dataset contains: - a voronoi.zip file with the scene and depth map, - a whites.zip file with the whites in png format, - a checkerboard.zip file with a rotating checkerboard in png format. - config_cam{1,2}.blend Blender files [3] for generating the Voronoi. References and links: [3] Blender Online Community, "Blender - a 3D modelling and rendering package." Blender Institute, Amsterdam: Blender Foundation, 2020. [4] K. Honauer, O. Johannsen, D. Kondermann, and B. Goldluecke, "A Dataset and Evaluation Methodology for Depth Estimation on 4D Light Fields" in Asian Conference on Computer Vision, 2016, https://github.com/lightfield-analysis/blender-addon https://github.com/dbonattoj/blender-addon [5] Perwass, Christian, et Lennart Wietzke. « Single Lens 3D-Camera with Extended Depth-of-Field », 829108. Burlingame, California, USA, 2012. https://doi.org/10.1117/12.909882.序列"ULB Voronoi - 模拟全光（Plenoptic）2.0"由比利时布鲁塞尔自由大学（Universite Libre de Bruxelles, ULB）布鲁塞尔高等理工学院（EPB）LISA实验室的Daniele Bonatto、Sarah Fachada、Gauthier Lafruit提供。 许可协议采用CC BY-NC-SA。使用本序列进行任何形式的出版或报告，均需引用如下文献。 [1] Daniele Bonatto、Sarah Fachada、Gauthier Lafruit，"ULB Voronoi - 模拟全光2.0"，2024。相关BibTeX条目：@misc{bonatto_voronoi_2024, title = {{ULB} {Voronoi} - {Simulated} {Plenoptic} {2.0}}, author = {Bonatto, Daniele and Fachada, Sarah and Lafruit, Gauthier}, month = feb, year = {2024}, doi = {10.5281/zenodo.10679358} } [2] D. Bonatto，《从多模态采集到照片级真实感视图合成——一种高质量实时多视图方法》，布鲁塞尔自由大学，2024。相关BibTeX条目：@thesis{bonatto_multi-modal_2024, location = {Brussels, Belgium}, title = {From multi-modal capture to photo-realistic view synthesis}, institution = {Universite Libre de Bruxelles}, type = {phdthesis}, author = {Bonatto, Daniele}, date = {2024-03}, } 本数据集由比利时布鲁塞尔自由大学布鲁塞尔高等理工学院LISA实验室图像合成与分析生产实验室制作。 数据集内容 本数据集包含利用Blender光场插件[4]构建的静态Voronoi场景。我们提供了方形棋盘格的标定图像，以及不同主镜头光圈下拍摄的白场图像，同时提供了该Voronoi数据集的Blender源文件。 本生成数据集遵循"每个微透镜等效于针孔相机"的原理。为此，我们借助Blender光场插件[4]生成了一组相机阵列，对基础场景进行拍摄，以此模拟全光相机的采集流程。 Voronoi场景包含两个带有Voronoi纹理的平面，分别距离微透镜阵列（micro-lens array, MLA）的距离为$|t_1|=2$mm与$|t_2|=2.5$mm。 为获取最大信息量，全光图像需采用典型的六边形布局[5]。但当前所用插件仅支持矩形布局，且要求相机在水平与垂直方向间距相等。而在六边形网格中，每行图像连续排布，垂直间距为水平间距的$sqrt{3}/2$倍。为弥补这一差异，我们首先创建了1×31的连续相机阵列，相机间距$B=0.2$mm，并在20帧的范围内按对应步长进行垂直偏移。为生成六边形布局中微透镜阵列间的行，我们对同一场景生成了第二组光场，偏移量为$(0.5, sqrt{3}/2)B$，最终每组光场共包含$20×31$个视角。 我们设置的相机参数如下：微透镜平行排布，焦距$f=100$mm，传感器尺寸为35mm，孔径直径$D=30$像素，基线间距$B=0.2$mm。上述参数对应全光相机的参数$s=frac{Bf}{ ext{sensor}}=0.5714$mm。 我们开发了脚本，利用两组光场生成六边形网格，并通过圆形掩膜生成模拟全光相机的图像。微图像的朝向决定了最终全光相机的配置：伽利略式（Galilean）或开普勒式（Keplerian），分别对应$t_1=±2$mm与$t_2=±2.5$mm。生成的伽利略式全光图像尺寸为$1095×945$。 本数据集包含以下文件： - voronoi.zip：包含场景与深度图 - whites.zip：包含PNG格式的白场图像 - checkerboard.zip：包含PNG格式的旋转棋盘格图像 - config_cam{1,2}.blend：用于生成Voronoi场景的Blender源文件[3] 参考文献与链接 [3] Blender在线社区，"Blender——一款3D建模与渲染软件包"，阿姆斯特丹：Blender基金会，2020。 [4] K. Honauer、O. Johannsen、D. Kondermann与B. Goldluecke，"面向4D光场的深度估计数据集与评估方法"，收录于亚洲计算机视觉大会，2016。相关链接：https://github.com/lightfield-analysis/blender-addon ；https://github.com/dbonattoj/blender-addon [5] Perwass, Christian, 与 Lennart Wietzke. "具有扩展景深的单镜头3D相机"，829108，美国加利福尼亚州伯灵格姆，2012。DOI: 10.1117/12.909882。