DeepStack On Raspberry PI + Intel Movidius Stick and Unlimited Free Activations


#1

Hello Everyone, thanks for all the feedback. feature requests and incredible support that has pushed us to always work on improving DeepStack.
I am excited to share that today, we have released the Alpha Version of DeepStack on the Raspberry PI with full support for the Intel Neural Compute Stick.

Also, we listened to your complaints on activations and now, with a free key, you can perform unlimited activations.

This represents a new phase for the development of DeepStack, over the next weeks, we shall be releasing new updates to the Docker and Windows Versions with improved stability, support for custom object detection, improved security and more!

Thank you all.


#2

Great work on this!
I want to confirm expected performance. Using the vision/detection model inference time is around 2.5 sec, is this typical?
Also I unplugged the NCS stick while the container was running and receive:

{'success': False, 'error': 'invalid image'}

on a request, I suggest

{'success': False, 'error': 'no NCS stick'}

might be better?`


#3

Hello @robmarkcole Thanks for trying it out. The expected speed for the detection model is about 2.5 secs.
On the error, you are absolutely right, we would definitely provide better error reporting in the next update. The beta version would offer more improvements and better experience than the current alpha version.


#4

If the processing speed is 2.5 second per image, is there method to bundle multiple computing stick on same RPI module?

2.5 second with detection feature is not very useful in practical environment.

Thank
Jerry


#5

Hi @JDU we plan to add support for multiple sticks in the beta versions of DeepStack. This would allow you process more images at once. The current version is an Alpha and we are working on further improvements in the beta.