/Local_Light_Field_Synthesis

Local Light Field Synthesis (Pratul P. Srinivasan, Tongzhou Wang, Ashwin Sreelal, Ravi Ramamoorthi, Ren Ng ICCV 2017)

Primary LanguageJupyter NotebookApache License 2.0Apache-2.0

Learning to Synthesize a 4D RGBD Light Field from a Single Image

Pratul P. Srinivasan, Tongzhou Wang, Ashwin Sreelal, Ravi Ramamoorthi, Ren Ng

In the International Conference on Computer Vision (ICCV) 2017 (Spotlight Oral Presentation)

Paper, Video, Flowers Dataset (~169 GB), Supplementary Material

Example Input 2D Image

Example Input 2D Image

Predicted 4D Ray Depths

Predicted 4D Ray Depths

Synthesized 4D Light Field

Synthesized 4D Light Field

Synthesized Synthetic Depth-of-Field (Focused on Flower)

Synthesized Synthetic Depth-of-Field

Synthesized Synthetic Depth-of-Field (Focused on Background)

Synthesized Synthetic Depth-of-Field

Contents

This repository contains:

  1. Local_Light_Field_Synthesis.ipynb Jupyter notebook with an implementation of our algorithm. Note that this code may contain slight updates and modifications to the code used in our paper.

Dependencies

This code depends on a working installation of Tensorflow and basic Python libraries (numpy, scipy, matplotlib).

Acknowledgments

This work was supported in part by ONR grants N00014152013 and N000141712687, NSF grant 1617234, NSF fellowship DGE 1106400, a Google Research Award, the UC San Diego Center for Visual Computing, and a generous GPU donation from NVIDIA.