Dense Image Over-Segmentation on a GPU

The goal of this project is to take such an algorithm and implement it on a GPU using CUDA in the hope of accelerating it such that it theory it could operate at interactive frame rates on moderately sized images. The TurboPixels[1] algorithm was chosen as the baseline, as it is faster than other algorithms such as Normalized Cuts[2] while providing as high or higher quality results. Figure 1 illustrates an example segmentation produced by TurboPixels.