Winograd-Based Real-Time Super-Resolution System on FPGA

Abstract

With the rapid development of computer vision theory and visual display devices, High Frame Rate (HFR) and Ultra High Definition (UHD) techniques have received increasing attention from academic and industry. As they put high demands on performance and energy-efficiency, efficient customized hardware is required. In this paper, we propose an FPGA-based super-resolution system that enables real-time UHD upscaling in both high image quality and high frame rates. Our system crops each frame into blocks, measures their total variation values, and dispatches them accordingly to a neural network or an interpolation module for upscaling. We also propose a fast transposed convolution algorithm based on Winograd algorithm, which reduces the number of multiplications. Experimental results show that the proposed super-resolution system achieves superior performance in both reconstruction performance and efficiency over previous works.