MEMOCODE 2013 hardware/software co-design contest: Stereo matching

The MEMOCODE 2013 design contest problem is stereo matching. Given a stereo image pair (i.e., a left image and a right image), the challenge is to infer the depth information (i.e., third dimension) for each pixel in the image utilizing belief propagation algorithm. Contestants were given a month to develop a fast and/or cost-effective system for stereo matching. From a total of eight participating teams, three winners were selected based on pure-performance and cost-adjusted-performance metrics, taking into consideration inference accuracy achieved by their design solutions. The winning teams used hierarchical belief propagation algorithm on Intel multicore systems, streaming tree-reweighted message passing (TRW-S) algorithm on Convey HC-1 FPGA platform, and tile-based belief propagation algorithm on Nvidia GTX 680 GPU platform.