scikit-video

Video Processing SciKit BETA

Video processing algorithms, including I/O, quality metrics, temporal filtering, motion/object detection, motion estimation...

This is intended as a companion to scikit-image, containing all the algorithms which deal with video. There is a certain degree of overlap between image and video algorithms, for example a PSNR quality metric could be applied to pairs of images or pairs of video frames just as well. However, other algorithms are video-specific, for example a temporal denoise. This is the future home of the video-specific algorithms, as well as some of the algorithms which are not strictly video specific but are usually seen in a video context.

This also has some overlap with OpenCV. Roughly, the algorithms implemented here would be easier to hack on, and more research-oriented. Rather than building on top of a C/C++ framework, this will stay Python all the way, using whichever combinaiton of Numba/Theano/etc seems best for performance. This should add flexibility and better future ability to use GPU compute.

The project milestones are roughly:

• Add skeleton project from scikit-example - DONE
• Add video I/O by wrapping ffmpeg/avconv (similar to kanryu/pipeffmpeg) - DONE
• Add video metrics (from aizvorski/video-quality) - DONE
• More contributions roll in :)