Mr Gordon Chung
Mr Yiu Kei Li
Mr Tim K. L. Wong
Mr Teddy H.Y. Wong
Mr Peng Luo
Dr Kwok Wai Hung
Ms Xue Jiao Liu
Mr Chao Shi
Objective: Deep Convolutional Neural Network (DCNN) is an extremely powerful artificial intelligent model. It can bring ultra-high quality and novel intelligent functions to image processing devices. The major obstacle of the DCNN’s commercialization is lack of dedicated hardware solution to handle the huge computations required. This project investigates high efficient hardware solutions of DCNN, (including algorithms to training/optimizing DCNNs and dedicated hardware architectures), for image processing devices. Another objective is to align potential customers’ interests for this technology so as to identify the direction and roadmap of mass adoption. Research Methodology: In this research, an optimization algorithm will be developed to reduce the complexity of DCNNs for image processing tasks. Software simulation will be utilized to identify the efficient hardware architecture. A down-scaled FPGA design for concept verification and image quality assessment will be done for the commercialization in next step. Impact and benefit: The project deliverables will be technical foundations for a follow-up Platform Project to realize the ASIC design and reference system for image processing applications including camera, TV, display, video production, medical imaging, and so on. The developed technology will significantly enhance core competencies of relevant players in the market.