Belief propagation on probabilistic graphical models such as Markov random fields is the basis for a variety of applications, especially applications in computer vision. In this project, we used Vivado-HLS, a C-to-gates framework to rapidly prototype and test a number of accelerators for belief propagation on an FPGA. We used a Xilinx Zynq platform and our experiments show that using a dedicated FPGA accelerator provides a 2x speedup compared to a processor-only approach. Further, we note that the speedup is limited by communication bandwidth, and that higher speedups could be achieved using a higher-bandwidth bus.