DeepStack for Raspberry Pi + Intel Neural Compute Acceleration

Moses Olafenwa
Published in
3 min readAug 5, 2019


DeepStack is now available for the Raspberry Pi with APIs accelerated using Intel Neural Compute Stick.

We are glad to announce today that DeepStack AI Server that offers image recognition and detection AI APIs fully offline and on-the-edge is now available on the Raspberry Pi with its APIs accelerated by the Intel Neural Compute Stick. This release of DeepStack Pi is to allow easy deployment of AI for Home Automation and Industrial IoT, with all the benefits that comes with the AI server which are:

  • 100% privacy
  • Unlimited APIs
  • Zero API calls
  • Fully offline-capable

DeepStack Pi currently supports the following AI APIs

  • Face Detection
  • Face Recognition
  • Object Detection
  • Scene Recognition

Support for deploying Custom Recognition APIs will be added soon. To use DeepStack on Raspberry Pi + Intel Neural Compute Stick, kindly follow the steps detailed below.

Step 1 — Install DeepStack Pi

The PI Version is in alpha and would be regularly improved for optimal performance.

The minimum system requirements are:

  • Raspberry PI 3B+
  • Intel Movidius Neural Compute Stick 2

Open the terminal in your Raspberry Pi and run the commands below.




Step 2 — Start and Activate DeepStack Pi

Once the installation is done, run the command below to start the Object Detection API.

N.B: Please ensure your Intel Neural Compute Stick is connected to your Raspberry Pi every time before running DeepStack Pi.

sudo deepstack start "VISION-DETECTION=True"

By default, DeepStack Pi will run on PORT 80.

Visit localhost:80 in your browser to load the DeepStack Pi local admin. The interface below will appear.

VOILA! You are now running AI APIs locally on your Raspberry Pi, accelerated by the Intel Neural Compute Stick.

Step 3 — Using the APIs
Next, you can put your locally running AI APIs to test by trying the simple object detection code provided below on the sample image below .

Sample Image

Sample Python code

Sample NodeJS code

Sample CSharp code

When your code runs, the result from your DeepStack Pi will be as below.

If we draw the generated bounding boxes on the test image, we will have the result below.

For full documentation for using DeepStack Pi, visit the documentations linked below.






Moses Olafenwa

Software Engineer @BabylonHealth, Prev. @Microsoft. A self-Taught computer programmer, Deep Learning, AI Engineer.