This repo allows you to run OpenVINO AI/ML demos on Intel x86 hardware like a NUC, in a container, managed by balenaCloud. You need a monitor, keyboard/mouse, and USB webcam hooked up to the NUC.
This is interesting for several reasons, but first and foremost it allows you to remotely update an Edge AI device and push new applications and models to a device no matter where it is located. In this repo, as a sample, we download and install Object Detection and Human Pose Estimation, along with their corresponding models: pedestrian-detection-adas-0002, product-detection-0001, and human-pose-estimation-0001.
You will need to go through the normal balena device deployment steps, as documented here: https://www.balena.io/docs/learn/getting-started/intel-nuc/python/
Once your device is online, this repo has been cloned and pushed to the device, and the containers are running, you can hot-swap among the application and the model via Service Variables in the balenaCloud dashboard:
You need to define 3 Service Variables: DEMO, MODEL, and ARCHITECTURE_TYPE
The possible values for DEMO are:
object_detection_demo
human_pose_estimation_demo
For the MODEL Service Variable, if you have selected Object Detection, you have a choice of two possible values to enter: pedestrian-detection-adas-0002 which will detect humans, or product-detection-0001 which will detect about a dozen or so common grocery items.
If you seledcted Human Pose Estimation, then you only have one choice, and you will need to enter human-pose-estimation-0001 for the Model.
Finally, the ARCHITECTURE_TYPE Variable, has only 1 possible candidate for each. If you selected Object Detection, you will need to enter ssd as the value, but if you selected Human Post Estimation then you need to enter openpose.
Once your 3 values are entered, your container will start, and you'll see the device do inferencing on the live camera feed!