Can NCS2 be used for running a program other than inferencing on Nueral Networks? Can I run a simple python program on NCS2, applying a Gaussian filter on an image?
I ask this because on the intel page for Movidius™ Myriad™ X, which is the VPU used in Intel Computer Neural Stick 2, under the 16 High Performance SHAVE Cores it is stated:
"These programmable processors, with an instruction set tailored for computer vision, can be used to run traditional computer vision workloads, or can complement the Neural Compute Engine by running custom layer types for CNN applications thanks to extensive support for sparse data-structures."
First one all, the processors are programmable, does that mean I can program them as I'd like?
Second, it's said that it can run TRADITIONAL CV workloads. What is that exactly? Does it include, say, running a simple filter on an image?
Thanks in advance for answering.
Thanks for reaching out.
The Neural Compute Stick (NCS2) basically used for OpenVINO inferencing and computer vision as well, but the OpenVINO library must be included. Therefore, OpenVINO does use the OpenCV library. If you are planning to run your own program in NCS2, it might not work well. The NCS2 is designed to work with the OpenVINO toolkit.
Thanks for the answer,
So I take it that NCS2 can be used for computer vision tasks such as applying a simple filter to an image. But as far as I know NCS2 can only be used as a plugin for OpenVINO inferencing. If it can be used in another way, is there documents about it?
We don’t have any documentation for that since it has not been tested or validated by our developer. You can try to check this Intel® Neural Compute Stick 2 and Open Source OpenVINO™ Toolkit to understand more regarding the features of the NCS2 and OpenVINO toolkit.
Thank you very much for the time,
But the link you provided doesn't seem to answer my question.
"Can NCS2 be used for applying a customized simple filter to an image?" If so, is there any official documents or any sample code on how to use NCS2 for something other than inferencing?
We cannot validate any customized model that has not been tested by our developer. As mentioned before, you could try but we cannot guarantee the performance for that. There are OpenVINO samples and demos that you use as your reference for your model. You can also check the supported network for MYRIAD Plugin to build your model:
This thread will no longer be monitored since we have provided a relevant explanation. If you need any additional information from Intel, please submit a new question.
Start here: in the GNA section of the properties list.
I get that Intel coo will try to get the development part ready for answers. While the Edge technology is just available through open library resources. So is the documentation. I'm really trying to explain like ExplaningComputers, but I can't.
I get what you are trying to do and it just seems impossible.
You may want to search out more libraries. Intuit they may be a few available. Don't forget you are trying to program the unit. It has no disk space and it actually has a pretty large ram section. As far as Edge technology is described you wouldn't find hardware with that would service your ideal programming environment.
From personal experience filter processing is 0-255 color space with x, y, z coordinates. Don't test that on me. The coordinates may be a, b, c. Or something else. Just look it up. The idea is the filter libraries for computers are larger than the OpenVino. You just wouldn't find a reasonable solution.
You can process filters on your computer, and with an additional program even if you write one yourself. That's not what you want to do.