London, England – 17th May, 2022 – Imagination Technologies and Visidon Oy partner to power the transition to deep-learning-based super-resolution for embedded applications across mobile, DTV, and automotive markets. This AI-based technology allows users to upscale low-resolution images and videos up to 4K and 8K, through advanced algorithms. The IMG Series4 NNA enables these demanding processes to be done in real-time with greater efficiency within tight power and thermal SoC budgets.
Imagination’s IMG Series4 neural network accelerator (NNA) brings high compute performance and leading power efficiency for advanced AI image processing through its Tensor Tiling technology. Combined with Visidon’s expertise in deep-learning video enhancement networks, the solution will give OEMs the ability to include leading-edge AI enhancement for applications ranging from image restoration and colourisation, to drastically reducing video streaming bandwidth by up-sampling low-resolution source content.
Matti Niskanen, CTO, Visidon, says, “Power efficiency and high performance are two of the main pillars that Visidon builds its solutions on. We were surprised to see our deep-learning networks run so easily on Imagination’s IMG Series4 NNA and we were impressed with the overall power efficiency maintained while processing high compute workloads. We look forward to our partnership evolving as we unlock the future of AI-powered super-resolution technology together.”
Gilberto Rodriguez, Director of Product Management, Imagination, says, “As two companies that build their technology on a performance and power efficiency foundation, we are excited to drive the development of advanced AI image processing together, creating opportunities for OEMs to combine leading IP and software to deliver super-resolution applications. We hope our collaboration will unlock new opportunities for innovation across mobile, DTV, and automotive SoCs.”
The IMG Series4 AI compute engine with Tensor Tiling technology has been designed to provide low system bandwidth and high inference rate execution of convolutional-based neural networks – the main algorithm found in Visidon’s set of super-resolution solutions.
Visidon has developed a set of deep-learning-based super-resolution networks that can upscale and remaster video streams up to 4K (2160p) and 8K (4320p) to fit modern display resolutions. Visidon super-resolution increases the video frame size while maintaining image details. The result is superior output quality which outperforms the industry standard upscaling algorithms. Visidon often combines super-resolution with noise reduction technology to eliminate noise increase while preserving real details and clarity.