Imagination and Baidu PaddlePaddle create open-source machine learning library for Model Zoo

“Hardware and Ecosystem Co-Creation Program” offers enhanced support for AI developers and SoC designers

“Hardware and Ecosystem Co-Creation Program” offers enhanced support for AI developers and SoC designers

London, England, 27th SeptemberImagination Technologies and PaddlePaddle announce a new, jointly developed Model Zoo library.

This significant milestone in their partnership covers several types of AI processing, including image classification, image segmentation, and object detection. These Model Zoo resources will offer enhanced support to AI chip designers focusing on consumer, automotive, data centre and desktop markets and are globally available as open source.

Supporting the launch, a joint workshop will be available on the 28th of September for developers and manufacturers looking to use the Model Zoo resources.

Yanjun Ma, General Manager of Baidu AI Platform and Ecosystem, says: “As a leading supplier of processing technology and IP, Imagination enabled Baidu Paddle to construct a highly efficient and flexible software stack solution. The solution is developed based on the deep learning open-source PaddlePaddle framework and Imagination’s heterogenous computing IP through co-designing and optimisation. The collaboration will expand to other areas including data centre, consumer and RISC-V.”

Wallace Pai, Chairman, Imagination China, says: “PaddlePaddle is a long-time partner of Imagination, and we share a common vision for the future. We want to ensure that developers and manufacturers have access to the right tools for AI innovation. Our co-created AI ecosystem will continue to leverage our advanced capabilities in AI computing, enabling more support for software and SoC creators in the industry.”

Imagination and Baidu PaddlePaddle continue their collaboration, combining the advanced algorithms and flexibility of PaddlePaddle’s software with the high-performance compute capability of Imagination’s heterogenous AI accelerator cores. Part of the “Hardware Ecosystem Co-Creation Program”, the model supports chip and application developers in creating fully-optimised solutions for a range of deep-learning-based implementations.

Based on the companies’ rich AI expertise, the joint program has accelerated the development of a variety of verified software models, including image classification (EfficientNet), image segmentation (HRNet) and object detection (PP -YOLOE).  The consolidated Model Zoo library is designed to maintain and manage PaddlePaddle models validated by Imagination’s Neural Compute SDK, including model performance evaluation and reference instances for model deployment. Imagination and Baidu Paddle will hold a Model Zoo Deployment Practical Workshop on 28th September, offering comprehensive insights into the new AI resources. The workshop will cover end-to-end workflow for the deployment of the PaddlePaddle model on Imagination hardware.

Get regular insights

Sign up to receive the latest news and product updates from Imagination straight to your inbox.