Published at SIGGRAPH Asia 2015.
Authors:
- Nicolas Bonneel - CNRS LIRIS
- James Tompkin - Harvard Paulson SEAS
- Kalyan Sunkavalli - Adobe
- Deqing Sun - NVIDIA
- Sylvain Paris - Adobe
- Hanspeter Pfister - Harvard Paulson SEAS
This code implements this paper. It takes a processed or stylized video, say by a per-frame image processing operation, plus the original input video, and attempts to create a temporally-consistent version of that video.
Video (YouTube link):
I modified the original project to compile on Linux Ubuntu 18.04.It doesn't use ffmpeg anymore and works only with frame sequences.
mkdir build;cd build
cmake ..
make
usage : blindconsistency path/to/inputimages path/to/processedimages fileextension lambdaT path/to/outputresult framestart frameend
blindconsistency ../data/old_man ../data/old_man_autocolor png 1.0 ../data/out 1 50
will use old_man.[0001-0050].png and old_man_autocolor.[0001-0050].png to generate out.[0001-0050].png
- you need Opencv to be installed.
- Barnes et al., PatchMatch, SIGGRAPH 2009
- CImg
No commercial licence is given whatsoever. This code is distributed purely for education and for research purposes.