ONNX and SAS Event Stream Processing: Computer Vision for Edge and Cloud
Streaming analytics is a term that refers to using powerful machine learning on streaming data that is always in motion to derive insights quickly and uncover patterns for faster decision making. Using advanced machine learning for computer vision is a problem made more difficult when you consider the various machine learning frameworks for training models and the challenges of deploying these machine learning models to the powerful accelerated GPU and CPU hardware options now available. It’s a challenge when considering where and how to deploy these models to put them to work.
Increasingly, Open Neural Network Exchange (ONNX) formats are used to help with portability of models across frameworks and is an open standard format to represent machine learning models. SAS Event Stream Processing natively integrates with ONNX format models and has been enhanced for ONNX Runtime to reduce the complexity and time needed to deploy your models in various hardware settings. SAS Event Stream Processing delivers the following new benefits to solve these challenges:
- Native ONNX format model support for data scientist using 3rd-party machine learning frameworks
- Simplified design to integrate ONNX Models within your streaming projects
- Streamlined deployment to the edge or cloud reducing the inherent complexities of GPU and CPU hardware using ONNX Runtime
In the video below you’ll see how SAS Event Stream Processing supports ONNX Runtime and simplifies use of ONNX format models to deliver powerful computer vision models like OpenPose and TinyYolo to edge and cloud runtime environments.
Also, please check out our ESP using ONNX tutorial on GitHub to get started quickly. You can access the latest GitHub tutorial at GitHub for ESP ONNX Runtime Tutorial.
More use cases, examples, and reference architectures for SAS Event Stream Processing and IoT at: Getting Started with SAS IoT analytics.
Contact Steve Sparano for questions and more information.
