This dataset was generated in order to test scene flow methods using light fields. It was submitted along with a sparse to dense interpolation scene flow method at ICIP 2019.
The Sintel light field dataset is composed of 2 video sequences generated with Blender. We have used the production files of the open source movie Sintel made by Ton Roosendaal and the Blender Foundation and have modified them in order to render an array of 3x3 views. The generated sequences are :
Similarly to the MPI-Sintel Flow dataset, we have modified the scenes to generate not only the ‘final’ render, but also a ‘clean’ render without lighting effects, motion blur, or semi-transparent objects. For all of these videos, there is 50 frames and each view of the light field has a 1024x436 resolution. These videos are presented hereafter for the central view :
For each of these light field video sequences, the angular size of the light field is 3 by 3. The extracted views are numbered in the downloadable zip files as depicted in the diagram below (e.g.: Frame_0011_View_01_00.png for the view coordinates (1, 0) at frame no. 11).
The following videos are made using the 9 views from the 35th frame. The order of display is a serpentine order from the left up corner to the right bottom corner:
With each views, a ground truth optical flow, a ground truth disparity map and an occlusion mask were computed. Optical flows and disparity maps are put in the same image encoded with the OpenEXR format. For example, "Frame_0011_View_01_00.exr" contains as first and second channels the optical flow of the view (1,0) at the 11th frame and as a third channel the disparity map of said view. Scene flow files are sorted in the "scene_flow" folder and occlusion masks are sorted in the "occlusion_mask" folder. Here are below, as a video, the different maps related to the central view for the 2 sequences:
This work has been supported by the EU H2020 Research and Innovation Programme under grant agreement No 694122 (ERC advanced grant CLIM).
All datasets and benchmarks on this page are copyright by us and published under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License. This means that you must attribute the work in the manner specified by the authors, you may not use this work for commercial purposes and if you alter, transform, or build upon this work, you may distribute the resulting work only under the same license.