swiftframes

Motion-compensated Frame Interpolation Algorithm using Features Matching

Warning :

1. swiftframes is still in an early stage and offers extremly bad performances, both because the main computations are being done in python (python is very slow) and because of a few other things that could be optimized.
2. I didn't work on it since June 2018, but I might go back to it when I'll have more free time as I'm willing to try to make a more performant version (in Rust or C++, using concurency and vectorialization and maybe even using the GPU or an FPGA) and improve the quality with specialized home made features and more optimizations that I have in mind.
3. If anyone is interested in helping me, don't hesitate to contact me.

Dependencies :

• Python 3
• Numpy
• OpenCV, preferably with contrib (pip install opencv-contrib-python), but contrib is not mandatory (only needed if you want to try with feature detector and/or descriptors that are only in contrib like sift and surf)
• ffmpeg (or gstreamer) for the video part

How to use :

You can use :

test.py provides some example code on how to interpolate from some png files.

testVideo.py can be used to generate a interpolated video with double the original framerate.

you can use testVideo.py like that :

python testVideo.py VideoFileName

Code structure :

the "main" driver functions are in test.py and testVideo.py, since I don't have a real main function yet. motion.py contains the motion detection algorithm, it uses object from motion_match.py. interpolation.py contains the current interpolation algorithm. motion_match and motion_features contain objects used in the process the util folder contains some utility functions I used.

detector-tests and matching-tests were only used to compare the quality of different features regarding