Based on CUDA Cuts Code NPPI library remove graphcut algorithm after CUDA version 7.0. This project aims to give a fast version of graphcut algorithm on CUDA.

1. optimaze cuda cuts algorihtms in the paper
2. support complicated graph structure (not only plain images)

The first line contains three numbers describe the data size: width W height H, number of labels L

flowing H lines contains W integers which describes the ground truth data. (Not used in this code)

flowing H L lines contains W integers between label vertex. Here L = 2, the data describes the values of edges between pixels and source vertex, pixels and target vertex.

flowing H lines contains W - 1 integers which describes the values of horizontal edges.

flowing H - 1 lines contains W integers which describes the values of vertical edges.