This respitory contains code from Light Field Depth Estimation on Off-the-Shelf Mobile GPU - CVPRW 2018.
Only core parts of the code are included (C and GPU code)
The synthetic light field image are obtained from HCI and real light field image are captured using Lytro illum and decoded using Light Field Toolbox. The light field are decoded into a .png format and cropped into a 5x5 angular resolution due to Android memory limitation. Samples of light field images used in the paper are included in the LightField folder.
The Android.mk and application.mk files are included to enable OpenCL on Android. The neccesarry OpenCL include and .so files are also included.
*The android .apk files are available by request.
| Libraries | Link |
|---|---|
| OpenCL | https://www.khronos.org/opencl/ |
| Android Studio SDK+NDK | https://developer.android.com/studio |
| Light Field Image | http://hci-lightfield.iwr.uni-heidelberg.de/ |
| Light Field Toolbox | https://dgd.vision/Tools/LFToolbox/ |
| Cameras | Link |
|---|---|
| Lytro Illum | https://en.wikipedia.org/wiki/Lytro |
| Samsung S7 | - |
python script that runs on cross-platform supporting pyopencl.
- numpy
- opencv-python
- matplotlib
- tqdm
- pyopencl
- Linux
pip install pyopencl
python lf_depth.py
MIT