The Video Segmentation Project has been a collaborative effort between Georgia Tech and Google Research to put a state of the art segmentation and annotation system online for other researchers to use. In addition to making this system available online, we have made all of our source code open source and therefore available for you to use if our system does not fulfill your specific needs.
The algorithm, alongside numerous significant improvements, is based loosely on:
Efficient Hierarchical Graph Based Video Segmentation
IEEE CVPR, 2010