ReliaSolve worked with Applied Quantum Technologies on a brief project from March-May 2016 to accelerate maximum-likelihood calculations to solve for the objects causing a specific pattern due to Bragg scattering in prototype X-ray screening machines for airport baggage.
The original OpenCV-based calculations were several orders of magnitude too slow to be useful in the field. Through the use of sparse-matrix GPU calculations (cublas was not fast enough), custom threads (OpenMP was not fast enough) and running on a 4-node cluster with 3 GPUs each, the calculation was sped up by a factor of over 8000x. This brought calculations taking several hours down to about a second. The resulting solver is superlinear with problem size (the speed-up factor increases as the problem size increases), largely due to network latencies when communicating partial results between iterations.
A Qt-based workflow application named “ScatterGather” was developed to harness the resulting parallel simulator. This provides a GUI interface for launching simulation jobs, uses OpenGL to render the resulting slice data, and uses QWT to show normalized per-energy plots of the results. This enabled on-time delivery of a complete tool that exceeded performance specifications, along with operations manual and complete installation instructions.