Synthesizing Light Field From a Single Image with Variable MPI and Two Network Fusion (SIGGRAPH Asia 2020)

This is the code for the SIGGRAPH Asia 2020 paper.
Paper page: https://people.engr.tamu.edu/nimak/Papers/SIGAsia2020_LF/index.html

Our Light Field Dataset

We introduce a large Light Field dataset which contains over 2,000 light fields on general scene. This dataset is captured around TAMU campus. https://drive.google.com/file/d/1oHywOJAAYm8YrSH5mDubjTAwN8jrCC9y/view?usp=sharing

Environment

This code has been tested under Windows 10, Python 3.7.7, CUDA 10.1
Required libraries:

matplotlib            3.2.1  
numpy                 1.16.6  
opencv-python         4.2.0.34  
Pillow                7.1.2  
scikit-image          0.17.2  
scikit-learn          0.23.1  
scipy                 1.4.1  
tensorboardX          2.0  
torch                 1.5.0+cu101  
torchfile             0.1.0  
torchvision           0.6.0+cu101  

Usage

1. Download the testting data at https://drive.google.com/drive/folders/1hsHruwHIEuVQWfSlrgOSGlPrruWHdwqm?usp=sharing
2. Extract the data under the data folder
3. Run:

   python light_field_synthesis.py

4. Check the result:

   cd results
   python npy2lf_video8.py

The code can synthesize from 2D image, or synthesize 15 by 15 light field with minor change.
To test on your data, please use the Deeplens model to predict the depth:
https://github.com/scott89/deeplens_eval

To train the model, please download:
Standford Light field dataset: http://lightfields.stanford.edu/mvlf/
Our Light field dataset: https://drive.google.com/file/d/1oHywOJAAYm8YrSH5mDubjTAwN8jrCC9y/view?usp=sharing

Citation

@inproceedings{Li2020LF,
    author = {Li, Qinbo and Khademi Kalantari, Nima},
    title = {Synthesizing Light Field From a Single Image with Variable MPI and Two Network Fusion},
    journal = {ACM Transactions on Graphics}, 
    volume = {39}, 
    number = {6}, 
    year = {2020}, 
    month = {12}, 
    doi = {10.1145/3414685.3417785} 
}