Welcom to our FPFR project! This project includes a simplified version of our program demonstrating how our approach generates a light field based on the sparsely sampled views. We strongly recommand that the readers follow the instructions below to synthesize a light field by yourself. The code source will be available soon.
Our project is implemented in Python with the deep learning framework tensorflow (version 1.13.1)
Before launching the demo code, please make sure that the following packages are correctly installed.
Check list: tensorflow,numpy,matplotlib
We offered several light field scenes to test our demo code, running the run_demo.sh script to launch our demo. The command in run_demo.sh is:
python demo.py --scene_name=stilllife --mode=FPFR* --data_type=synthetic --angular_resolution=7 --inter_extra=inter
- scene_name is the name of light field scene.
- mode offers two mode, FPFR refers to a simple prediction, and FPFR* refers to the average of several predictions (more details in our paper).
- data_type indicates the type of light field data (synthetic or lytro).
- inter_extra indicates a demo for view interpolation or extrapolation (inter or extra).
- angular_resolution is the angular resolution of the generated light field, in the inter mode, angular resolution should be an integer greater or equal to 3; in the extra mode, angular resolution should be an odd integer greater than 3.
We offered several test scenes in the folder scenes, if you successfully launched our program, bravo! The synthesized light field can be found in the folder results.
Thank you for your attention, we sincerely hope you enjoy our project.