A GPU implementation of tiled belief propagation on Markov Random Fields

In the MEMOCODE Design Contest 2013, we are participating with a parallelized version of tiled belief propagation method for stereo matching. The proposed algorithm is implemented in CUDA programming model to leverage parallel processing capabilities of GPUs. In our solution, the original tiled belief propagation algorithm is combined with a number of novel optimizations specific to parallel programs in CUDA. For the given test inputs, the proposed solution runs in 7.96 milliseconds on Nvidia Tesla C2050, achieving acceptable accuracy with respect to the reference code. To the best of authors´ knowledge, there is no prior work in optimizing a parallelized version of the tiled belief propagation algorithm.