IT 4D Light Field (LF) dataset

About:

The IT light field dataset contains several synthetic 4D LFs including non-Lambertian objects and complex texture regions that mimic real images.

This is the first sparse 4D LFs dataset that includes ground truth segmentation label images, disparity, and depth maps for all LF views.

This dataset has disparity values between adjacent views within the range [-125,125] and consists of 11 4D LFs with (9×9) angular resolution and either (512×512) or (1280×720) spatial resolution. Our dataset contains several objects and challenging regions for segmentation, for example, non-Lambertian objects (e.g., glass and metal), complex textures, uneven lighting and overlapping objects with similar colors. As such, it can be used to evaluate various LF applications.

The dataset is generated by using LF Blender add-on [1], [2], and contained available free 3D models [3]–[6]. All generated LFs are rendered using Cycles rendering.

Details:

The database is organized in the following hierarchical folder structure as shown in Figure 1:

· The main directory, “IT light field dataset” contains 11 folders corresponding to each of the 11 light fields, namely: antique, antique2, balloons, breakfast, car, chess, garage, kitchen, leisure, room.

· Each of these 11 folders contains:

§ A folder for the Sub Aperture Images named (SAI): this folder contains 81 RGB light field views with .png extension.

§ Disparity maps file named (disparity): this file includes disparity maps for all 81 light field views with .mat extension. The data inside the file is organized as (height, width, rows of views, columns of views).

§ Depth maps file named (depth): this file includes depth maps for all 81 light field views with .mat extension. The data inside the file is organized as (height, width, rows of views, columns of views).

§ Segmentation labels file named (labels): this file includes labels maps for all 81 light field views with .mat extension. The data inside the file is organized as (height, width, rows of views, columns of views).

§ Thumbnails for the central view disparity, depth and labels images with .png extension.

§ A parameters file named (parameters): this file is generated by Blender and includes all camera parameters with .txt extension.

Figure 1: Folders hierarchy

Download:

The whole dataset can be downloaded from here as single zip file. If you prefer, individual light fields can be downloaded below. A brief description of individual light fields is available here.

· Antique

· Antique2

· Balloons

· Bedroom

· Breakfast

· Car

· Chess

· Garage

· Kitchen

· Leisure

· Room

How to reference the dataset:

If you use this dataset in your work, please cite our paper:

· M. Hamad, C. Conti, P. Nunes, and L. D. Soares, “Hyperpixels: Flexible 4D Over-segmentation for Dense and Sparse 4D Light Fields.” IEEE Transactions on Image Processing, vol. 32, pp. 3790-3805, 2023, doi: 10.1109/TIP.2023.3290523.

Contacts:

Feedback is welcome. Please provide your comments and suggestions by e-mail.

References:

[1] B. Foundation, “blender.org - Home of the Blender project - Free and Open 3D Creation Software,” blender.org. 2018.

[2] K. Honauer, O. Johannsen, D. Kondermann, and B. Goldluecke, “A Dataset and Evaluation Methodology for Depth Estimation on 4D Light Fields,” in Computer Vision - ACCV 2016 : 13th Asian Conference on Computer Vision, Taipei, Taiwan, November 20-24, 2016, Revised Selected Papers, Part III, S.-H. Lai, Ed. Cham: Springer, 2017, pp. 19–34.

[3] “Chocofur Main Page.” [Online]. Available: https://chocofur.com/. [Accessed: 09-Dec-2021].

[4] “BlenderKit - Get free 3D models, materials & more directly in Blender.” [Online]. Available: https://www.blenderkit.com/. [Accessed: 09-Dec-2021].

[5] “3D Models for Professionals :: TurboSquid.” [Online]. Available: https://www.turbosquid.com/. [Accessed: 09-Dec-2021].

[6] “Blender Free 3D Models - .blend download - Free3D.” [Online]. Available: https://free3d.com/3d-models/blender. [Accessed: 07-Feb-2022].