Light-Field Intrinsic Dataset, BMVC 2018

Abstract

Light-field imaging has various advantages over the traditional 2D photography, such as depth estimation and occlusion detection, which can aid intrinsic decomposition. The extracted intrinsic layers enable multiple applications, such as light-field appearance editing. However, the current light-field intrinsic decomposition techniques primarily resort to qualitative comparisons, due to lack of ground-truth data. In this work, we address this problem by providing intrinsic dataset for real world and synthetic 4D and 3D (only horizontal parallax) light fields. The ground-truth intrinsic data comprises albedo, shading and specularity layers for all sub-aperture images. In case of synthetic data, we also provide ground-truth depth, normals, and further decomposition of shading into direct and indirect components. For real-world data acquisition, we make use of custom hardware and 3D printed objects, assuring precision during multi-pass capturing. We also perform, qualitative and quantitative, comparison of existing intrinsic decomposition algorithms for single image, video, and light field. To the best of our knowledge, this is the first such dataset for light fields, which is also applicable for single image, multi-view stereo, and video.

Results

1. Synthetic Dense Light-Field Intrinsics



2. Synthetic Sparse Light-Field Intrinsics



3. Real-world 3D Light-Field Intrinsics



4. Real-world 4D Light-Field Intrinsics

Citation

Sumit Shekhar, Shida Beigpour, Matthias Ziegler, Michał Chwesiuk, Dawid Paleń, Karol Myszkowski, Joachim Keinert, Radosław Mantiuk, Piotr Didyk
Light-Field Intrinsic Dataset
British Machine Vision Conference (BMVC), 2018

@inproceedings{DBLP:conf/bmvc/ShekharBZCPMKMD18,
  author    = {Sumit Shekhar and
               Shida Beigpour and
               Matthias Ziegler and
               Michał Chwesiuk and
               Dawid Paleń and
               Karol Myszkowski and
               Joachim Keinert and
               Radosław Mantiuk and
               Piotr Didyk},
  title     = {Light-Field Intrinsic Dataset},
  booktitle = {British Machine Vision Conference 2018, {BMVC} 2018, Northumbria University,
               Newcastle, UK, September 3-6, 2018},
  pages     = {120},
  year      = {2018},
  crossref  = {DBLP:conf/bmvc/2018},
  url       = {http://bmvc2018.org/contents/papers/0431.pdf},
  timestamp = {Mon, 17 Sep 2018 15:40:26 +0200},
  biburl    = {https://dblp.org/rec/bib/conf/bmvc/ShekharBZCPMKMD18},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

© 2018 The Authors. This is the author's version of the work. It is posted here for your personal use. Not for redistribution.

Downloads

Related Projects

1. Towards a Quality Metric for Dense Light Fields


2. Efficient Multi-image Correspondences for On-line Light Field Video Processing


3. Light-Field Appearance Editing Based on Intrinsic Decomposition

Acknowledgements

We would like to thank Michal Piovarči for his help with the 3D printing. We would like to thank Anna Alperovich and Ole Johansen for valuable discussion. We would like to thank Anna Alperovich, Abhimitra Meka, and Elena Garces for kindly providing the necessary comparisons. We thank the reviewers for their insightful comments. The project was supported by the Fraunhofer and Max Planck cooperation program within the German pact for research and innovation (PFI).