2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (2008)
Anchorage, AK, USA
June 23, 2008 to June 28, 2008
P. J. Narayanan , Centre for Visual Information Technology, International Institute of Information Technology, Hyderabad, 500032. India
Vibhav Vineet , Centre for Visual Information Technology, International Institute of Information Technology, Hyderabad, 500032. India
Graph cuts has become a powerful and popular optimization tool for energies defined over an MRF and have found applications in image segmentation, stereo vision, image restoration, etc. The maxflow/mincut algorithm to compute graph-cuts is computationally heavy. The best-reported implementation of graph cuts takes over 100 milliseconds even on images of size 640×480 and cannot be used for real-time applications or when iterated applications are needed. The commodity Graphics Processor Unit (GPU) has emerged as an economical and fast computation co-processor recently. In this paper, we present an implementation of the push-relabel algorithm for graph cuts on the GPU. We can perform over 60 graph cuts per second on 1024×1024 images and over 150 graph cuts per second on 640×480 images on an Nvidia 8800 GTX. The time for each complete graph-cut is about 1 millisecond when only a few weights change from the previous graph, as on dynamic graphs resulting from videos. The CUDA code with a well-defined interface can be downloaded for anyone’s use.
P. J. Narayanan, Vibhav Vineet, "CUDA cuts: Fast graph cuts on the GPU", 2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, vol. 00, no. , pp. 1-8, 2008, doi:10.1109/CVPRW.2008.4563095