Where-what network with CUDA: General object recognition and location in complex backgrounds

Abstract

An effective framework for general object recognition and localization from complex backgrounds had not been found till the brain-inspired Where-What Network (WWN) series by Weng and coworkers. This paper reports two advances along this line. One is the automatic adaptation of the receptive field of each neuron to disregard input dimensions that arise from backgrounds but without a handcrafted object model, since the initial hexagonal receptive field does not fit well the contour of the automatically assigned object view. The other is the hierarchical parallelization technique and its implementation on the GPU-based accelerator using the CUDA parallel language. The experimental results showed that automatic adaptation of the receptive fields led to improvements in the recognition rate. The hierarchical parallelization technique has achieved a speedup of 16 times compared to the C program. This speed-up was employed on the Haibao Robot displayed at the World Expo, Shanghai 2010. © 2011 Springer-Verlag.