IBM Edge Application Manager (IEAM)

Edge deployment of container workload and ML model

Machine Inferencing and Object Detection examples using TensorFlow Lite, OpenVINO, pyTorch and MVI frameworks. Implemented with IEAM + MMS + OpenCV + Python

Introduction

This python based example implementation uses multiple containers and can be deployed on Intel NUC(amd64), Jetson Nano(arm64) and RaspberryPi 4(arm32) using following frameworks. The end-to-end deployment of containerized services via IEAM is the focus of these examples NOT the accuracy or performance of the ML model, though occasionally that is mentioned.

• MVI (IBM)
• Tensorflow Lite
• OpenVINO
• PyTorch

High level features:

• Object detection using various frameworks
• OpenCV based image capture and annotation
• MJPEG based streaming available on http:edge-device-ip-address:5000
• A simple Web UI to interactivly upload config using MMS to edge nodes.
• Intel Neural Compute Stick 2 (NCS2)
• Movidius MyriadX VPU
• HTTP and kafka message bridge

Publish

Development of containers, services, policies and corresponding defintion files. See publish directory.

Register

Instructions to register an edge device node to detect objects in a video stream See register directory.

Research, reference and acknowledgements

https://opencv.org

https://www.tensorflow.org/lite

https://github.com/EdjeElectronics/TensorFlow-Object-Detection-on-the-Raspberry-Pi

https://stackoverflow.com/questions/tagged/tensorflow

OpenVINO