Imagination, together with Baidu PaddlePaddle, created the Model Zoo open-source machine learning model library “Hardware Ecology Co creation Plan” to provide enhanced support for AI developers and SoC designers

Beijing, China – On September 27, Imagination and Feichuan jointly released the new Model Zoo model library. This important milestone of the cooperation between the two sides covers a variety of artificial intelligence processing technologies, including image classification, image segmentation and object detection. These Model Zoo resources will provide support for AI chip designers and system manufacturers focusing on the consumer, automotive and desktop server markets, and will be provided in open source worldwide.
In order to support this release, the two sides will hold a seminar on “PowerVR+Propeller Model Zoo Model Deployment Practice” for developers and system application manufacturers who want to use Model Zoo resources on September 28.
Ma Yanjun, general manager of Baidu AI Technology Ecology, said: “As a world-renowned processor technology and IP provider, Imagination has built efficient and flexible software stack solutions through joint definition and optimization based on the open source framework and Imagination heterogeneous computing IP technology of deep learning by flying oars. In the future, the cooperation between the two sides will gradually expand to data centers, consumption, Risc-V and other application fields.”
Bai Nong, director of Imagination China, said, “Baidu PaddlePaddle is a long-term partner of Imagination, and we share a common vision for the future. We hope to ensure that developers and manufacturers can obtain the correct AI innovation tools. The AI ecosystem we jointly created will continue to use our advanced capabilities in AI computing to provide more support for software and SoC creators in the industry.”
Imagination and Flying Propeller continue to work together to combine the advanced algorithm and flexibility of Flying Propeller software with the high-performance computing capability of Imagination heterogeneous AI accelerator core. As part of the results of the “hardware ecological co-creation plan”, the model supports chip and application developers to create fully optimized solutions for a series of implementation based on deep learning.
Based on the resources accumulated by the two sides in the field of artificial intelligence and other fields over the years, the cooperation of the hardware ecological co-creation plan has made rapid progress and rapidly formed a number of verified and available models, including the EfficientNet model of image classification task, the HRNet model of image segmentation task, and the PP-YOLOE model of object detection task. The purpose of Model Zoo is to maintain and manage the propeller model verified by the Imagination Neural Network Computing SDK, including reference examples of model performance evaluation and model deployment.
The “PowerVR+Flying Propeller Model Zoo Model Deployment Practice” seminar held by Imagination and Flying Propeller on September 28 will provide new resources for comprehensive understanding of AI, covering the end-to-end workflow of deploying flying propeller models on the Imagination hardware.
About Imaging Technologies
Imagination is a company headquartered in the United Kingdom, committed to building semiconductor and software intellectual property (IP), so that customers can gain sufficient advantages in the competitive global technology market. The company’s graphics processor (GPU), central processing unit (CPU) and artificial intelligence (AI) technology can achieve outstanding PPA (power consumption, performance and area) indicators, fast time to market and lower total cost of ownership (TCO). Imagination IP-based products are used by billions of people around the world in their mobile phones, cars, homes and workplaces.
About Baidu PaddlePaddle
Based on Baidu’s deep learning technology research and business application for many years, Feichuan is the first independent, functional, open source and industrial-level deep learning platform in China, integrating deep learning core training and reasoning framework, basic model library, end-to-end development kit, and rich tool components. At present, Feichuan has gathered 4.77 million developers, created 560000 models based on Feichuan’s open source deep learning platform, and served 180000 enterprises and institutions. Flying oars help developers quickly realize AI ideas, innovate AI applications, and support more and more industries to realize industrial intelligent upgrading as a basic platform.