When I try to execute my model in Raspberry PI using NCS2, It's too slow.

[Lee__Hanbeen]

Hi. I am a student studying deep running in Korea.

I recently got NCS2 and wanted to use it.

I visited this site[Here] to use the Neural Computer Stick 2 in Raspberry PI.

I used this site[Here] to replace my ckpt model with IR (.xml, .bin).

I successfully changed ckpt to pb file using graph_io and changed this file to .xml, .bin file using model_optimizer.

Although there were some difficult points, I succeeded in converting smoothly.

The problem occurred when creating and running the source on Raspberry PI.

here (although this is written by Korean, you just look at bottom of page) is my source. 

It works well.

BUT!! It's very very slow. It takes about 4.7 seconds to deduce a single image.

A network that is smaller than this network is taking too long and can not infer reasonably well after 10 minutes.

My guess is that Raspberry PI is not recognizing the NCS2 device, or there might be a configuration problem.
Which part should I check?

[Olivero__Alberto]

What Raspberry do you use ?

Is it Raspberry 2B or 3B or 3B+ ?

[Lee__Hanbeen]

I use Raspberry PI 3 b model.

Is there a good way to solve this?

[Lee__Hanbeen]

Resolved.
Instead of using IE as a backend in OpenCV,
Using IE directly, the inference time was shortened from 4.7 seconds to 0.01 seconds.
But there is still a problem. The inference for color images (128, 64) is normal, while the grayscale image is still ending at the end of infinite time.

I have written the relevant source code on my GITHUB.here
It is in Korean, but you can see only the source at the bottom.

[Olivero__Alberto]

Multiple peoples report issues using P3B or P3B+ in the following, suggesting potential solutions.

https://software.intel.com/en-us/forums/computer-vision/topic/800823

Intel report they tested only on armv7 means P2B.