Approximate Computing for Motion Picture Camera Processing
Trading off resources (energy, area, performance)
against application quality in a controlled way
Topic Introduction
In application domains like image processing or data analysis, ever-increasing performance demands push the capabilities of computational systems to their limits. This is especially true for professional imaging systems such as motion picture cameras that have to handle growing resolution, frame-rate and dynamic range. With technology scaling plateauing out, engineers are forced to rethink their approach to system design. The research field of approximate computing provides a new design paradigm which trades off accuracy against computational resources. The main challenge when applying this idea to real problems is the characterization and optimization of benefits while ensuring that quality levels do not fall short of certain acceptability standards.
This research project focusses on the application and optimization of approximation methods in professional image processing systems such as digital motion picture cameras (ARRI) and laser scanners (SmartRay) which are based on FPGA devices. We intend to use a SoC-based prototying system to characterize suitable approximation methods and we plan to utilize machine learning methods (like rule-based systems) to optimize approximation parameters.
Research questions
How to approximate? - Which approximation methods are suitable and benefitial for image processing on FPGA? - How can we model their benefits and errors accurately and efficiently?
Where to approximate? - How can we find an optimal configuration of approximation parameters? - Which quality-energy trade-off knobs are useable in FPGA design flows?
What is the target quality (bound)? - Which quality metrics are suitable for the application and useable in the system optimization? - Which threshold can be tolerated?
Publications
2019 Simon Conrady, Manu Manuel, Arne Kreddig, Walter Stechele: LCS-Based Automatic Configuration of Approximate Computing Parameters for FPGA System Designs. Genetic and Evolutionary Computation Conference Companion (GECCO ’19 Companion), 2019 (accepted) (Link)