ONNX support in the latest CDNN compiler
Neural Networks Can Be Applied to CEVA-XM Vision DSPs
Ensuring an open, interoperable AI ecosystem
CEVA today announced that it is adding support for the Open Neural Network Exchange (ONNX) format to its latest CEVA Deep Neural Network (CDNN) compiler.

ONNX is an open standard for sharing deep learning models across platforms and services. An open format created by Facebook, Microsoft, and AWS to promote interoperability and computer software portability within the AI community, allowing developers to use the right mix of tools for the project at hand without being tied to a specific framework or ecosystem.

It also ensures interoperability between different deep learning frameworks, allowing developers to train neural networks using any machine learning framework and leverage and apply other AI frameworks.

With CDNN supporting ONNX, developers can now use any ONNX-compatible framework and import the generated models to deploy on CEVA-XM vision DSPs and NeuPro AI processors.

Ilan Yona, vice president and general manager, Vision Business Unit at CEVA, said: “CEVA wants to ensure that AI application developers can leverage the capabilities and ease of use of a wide variety of deep learning frameworks while still being able to best fit their specific use cases, while fully supporting an open and interoperable AI ecosystem.”

Additionally, he added, “By adding ONNX support to our CDNN compiler technology, we are providing CEVA-XM and NeuPro customers and ecosystem partners with a much wider range of capabilities to train and power neural network-based applications.”