EVA introduction
  • 13 Apr 2022
  • 2 Minutes to read
  • Contributors
  • Dark

EVA introduction

  • Dark

Article Summary

EVA SDK is a unified edge vision analytics service-ready software platform that enables ADLINK AI hardware, making it easier for users to develop optimized edge AI vision applications by simplifying integration and focusing on essential functionality. Users can leverage readyto- use open-source plugins to facilitate each stage the AI vision project lifecycle, including image capture and processing, AI inference, postprocessing, and analytics. This "One API" framework allows users successfully to build a proof-of-concept in two weeks and speed up mass deployment time.


EVA is leveraging the opensource community - Gstreamer as the foundation of SDK. Users can fully utilize the Gstreamer resource and integrate them with EVA.
In addition, EVA provide more featuares to reduce the complexity when develop the AI application at the edge.

AI Pipeline Studio:

EVA IDE support the low code/ no code programining. User can construct the application as a pipeline by drag & drop the elements from image ingestion, image processing, AI inference and result ouput. The IDE support the runtime debug, what you see is what you get.

Abundant Ready-to-use Plugin:

EVA integrated and verifed abundant Gstreamer plugins that is suitable for the vision analytics, like varity of image src, image processing, AI inference, image output plugins. EVA also provide the python bounding, that user can use python to implement an new plugin and run with C plugin without any problem.

Optimized AI model:

In order to support most vision AI use cases, EVA support the optimzed CNN AI model on classfication, object detection, segmentation and human pose detection. The details list could be found on "Support Models" session

Support Hybird AI Inference Engine:

The most powerfull features of EVA is that it support the multiple inference engine, currently it support the OpenVINO, TensorRT and ONNX runtime. The Huawei MindX and Qualcomm SNPE are in the roadmap. By integrating the vendors SDK, EVA provide the unifed interface that user can easily switch the inference engines without and code migration or porting needed. For example, user could switch from TensoRT to OpenVINO inference just seconds. That may required months effort in the pass. Moreover, user can run them both to fully utilize the system resource.

Support Multiple Industrial Cameras:

A robst image src is the key for the edge AI infernece. Without a stable image ingestion on all kind of environment, AI can't get the quality result. EVA support most of popular industrial camera vendors and interface. By leverage the EVA provided plugin, user can switch between differnet cameras without coding and select the best camera for their edge environment.

Support the Windows and Ubuntu Operation System:

Since most of traditional vision application still runing on windows, EVA is designed as operation system independent. User can run the EVA pipeline on both windows and linux system. Make it easy to integrate with user's legacy application and develop the AI with user's familar tool kit.
[EVA in Windows]

[EVA in Linux]

Was this article helpful?

What's Next