1. Begin with a real-time detector starter app and model set
For this example, we will use the alwaysAI real-time object detector starter app, which utilizes the MobileNet SSD model. If you haven't downloaded the alwaysAI starter applications, do so now by entering 'aai get-starter-apps' from your terminal. This will download the starter applications into the current working directory. Navigate into the 'realtime_object_detector' directory.
Here, our object detector app is detecting a potted plant in our office. Note that the inference time of the device without the accelerator is .710 seconds.
2. Change the object detection engine
To use the accelerator, change the DNN to DNN OpenVINO.
3. Re-deploy and run the start command
Then re-deploy the app using alwaysai app deploy and re-start it using the alwaysai app start command.
4. Check your new inference time
Double-check the inference time from your object detector app in the Streamer on your browser.
You can see that after using Intel’s Neural Compute Stick 2, the inference time has dropped to .094 seconds. That's a significant improvement for this edge device.
Get started now
We provide developers with a simple and easy-to-use platform to build and deploy computer vision applications on edge devices.