Light Field Reconstruction from Coded images

• Official Tensorflow implementation for our TCI paper on A Unified Learning-Based Framework for Light Field Reconstruction From Coded Projections

Demo Video

Prerequisites

• NVIDIA GPU + CUDA + CuDNN
• Python 3.6
• tensorflow-gpu
• Other Python dependencies: numpy, scipy, scikit-image, imageio, matplotlib

Datasets

Kalantari et al.: http://cseweb.ucsd.edu/~viscomp/projects/LF/papers/SIGASIA16/

Flowers dataset: https://cseweb.ucsd.edu/~viscomp/projects/LF/papers/ICCV17/lfsyn/LF_Flowers_Dataset.tar.gz

Training:

Run nbs/ulf_focdef_train.ipynb

Testing:

Run nbs/ulf_focdef_test.ipynb

Pretrained models:

https://drive.google.com/drive/folders/1Ja2bFg9ThpjgRHw2Wg0gNubdKvOlJA6q?usp=sharing

Citation

@ARTICLE{8878019, 
         author={A. K. {Vadathya} and S. {Girish} and K. {Mitra}}, 
         journal={IEEE Transactions on Computational Imaging}, 
         title={A Unified Learning-Based Framework for Light Field Reconstruction From Coded Projections}, 
         year={2020}, 
         volume={6}, 
         number={}, 
         pages={304-316},
}