Accelerated gSLIC for Superpixel Generation used in Object Segmentation

Accelerated gSLIC for Superpixel Generation used in Object Segmentation

Robert Birkus

Abstract. The goal of our work is to create a robust object segmentation method which is based on superpixels and will be able to run in real-time applications.

The SLIC algorithm proposed by Achanta et al. [1] is a superpixel segmentation algorithm based on k-means clustering, which efficiently generates superpixels. It seems to be a good trade-off between the time consumption and robustness. Important advancement towards the real time applications using superpixels has been proposed by the authors of the gSLIC – a modified SLIC implementation on the GPU (Graphics Processing Unit) [2].

In this paper, we present a significant acceleration of this superpixel segmentation algorithm gSLIC implemented for the GPU. A different strategy of the implementation on the GPU speeds up the calculation time twice and more over the presented GPU implementation. This implementation can work in real-time even for high resolution images. We also present our method for merging of similar superpixels. This method uses an adaptive decision procedure for merging of superpixels. Accelerated gSLIC is the first part of this proposed object segmentation method.


[1] R. Achanta, A. Shaji, K. Smith, A. Lucchi, P. Fua, and S. Susstrunk. Slic superpixels. Technical report, Ecole Polytechnique Fedralede Lausanne , Report No. EPFL-REPORT-149300, 2010.
[2] C. Y. Ren and I. Reid. gSLIC: a real-time implementation of SLIC superpixel segmentation. Technical report, University of Oxford, Department of Engineering, Technical Report (2011)., 2011.

Paper is avaible at CESCG proceedings:

Source code:

Solution (Visual Studio 2012, V11):