NCS2 and Raspberry Pi 4

[Mehta__Jay]

Can any one share the information on compatibility of NCS2 stick with Raspberry Pi 4? I am looking to start a new prototyping with Raspberry Pi 4.

Thanks.

[JesusE_Intel]

Hi Jay,

Thanks for reaching out! The Intel Neural Compute Stick 2 has only been validated on the Raspberry Pi 3. We are currently working on adding support for the Raspberry Pi 4 but I don't have an ETA.

However, take a look at the following discussion where a user has instructions for using the Intel NCS 2 with the Raspberry Pi 4. I've not test this myself so it may or may not work. 

https://software.intel.com/es-es/node/813265

Hope this helps!

Regards,

Jesus

[RTasa]

Hi Jesus, Any progress on the NCS2 , Pi 4 validation?

[JesusE_Intel]

Hi BobT,

Yes, we actually posted instructions to setup the Raspberry Pi 4 with the Intel Neural Compute Stick 2. Give it a try!

https://software.intel.com/en-us/articles/raspberry-pi-4-and-intel-neural-compute-stick-2-setup

Regards,

Jesus

[RTasa]

So far no go. From a fresh install after waiting hours for a compile for the first part OpenCV failed to make and install correctly. The test failed. And did you notice this was for Arm7 ? Pi 4 is Arm8 https://software.intel.com/en-us/articles/ARM-sbc-and-NCS2 hoping for an update soon.

[JesusE_Intel]

Hi Bob,

The instructions I provided above are for building the open-sourced version of the OpenVINO toolkit and should work on both the Raspberry Pi 3 and 4. The instructions are for Armv7, however, Raspbian is built for Armv7 to maintain compatibility. I will go though the built process with my Raspberry Pi 4 today and get back to you.

By the way, the OpenVINO toolkit also has a Raspbian OS package that was pre-built with 2019 R2. If you don't want to build from source, you can download the package below and follow the Install OpenVINO™ toolkit for Raspbian* OS documentation. 

https://download.01.org/opencv/2019/openvinotoolkit/R2/

Regards,

Jesus

[Cairns__Jim]

Hi Jesus - Do you have any performance benchmarks for the RPi4 and NCS2?  We are running a portable object counting solution (counting pigs) using an Atomic Pi and NCS2 and SSD model.  Ideally we'd like to run an R-CNN for better accuracy but the fps have been far too low with the Atomic Pi.  Any insights would be great.

Thanks - Jim

[JesusE_Intel]

Hi Jim,

You should be able to run the benchmark app included in the OpenVINO toolkit with your model. Take a look at the the benchmark app documentation: https://docs.openvinotoolkit.org/latest/_inference_engine_samples_benchmark_app_README.html

Regards,

Jesus

[landau__cyrill]

Hey Jesus and other intel workers,

   None of the stuff seems to work. I am not experienced in linux and am programming a raspberry pi for motion sensing and analysis with a neural compute stick. all my previous development work was in windows. what is this? nothing is working. is the neural compute stick janky?

[JesusE_Intel]

Hi cyrill,

Could you please start a new discussion and provide additional information of the issue you are running into and what commands you used?

Also, make sure you are using the latest OpenVINO toolkit for Raspbian OS package (2020.1) and followed the installation guide.

Regards,

Jesus