I have been trying to install openVino 2021.4 on my new Nvidia Jetson nano 2GB Developer Kit with the following system installed:
VERSION="18.04.5 LTS (Bionic Beaver)"
PRETTY_NAME="Ubuntu 18.04.5 LTS"
as well as :
Operating System: Ubuntu 18.04.5 LTS
Kernel: Linux 4.9.201-tegra
When I try run the Install_GUI.sh I get the following errors and cant seem to find a solution:
./install.sh: 60: [: -lt: unexpected operator
./install.sh: 63: [: -eq: unexpected operator
The IA-32 architecture host installation is no longer supported.
The product cannot be installed on this system.
Please refer to product documentation for more information.
I bought the Nvidia jetson specifically to run the OpenVino on it for people tracking etc.
Thanks for reaching out. I suspect the error arises due to the OpenVINO toolkit requires an x86-64 host computer with Ubuntu 18.04 (LTS). You could try to build the OpenVINO toolkit from the source as an alternative and see if this works. Refer to this Github page for the instruction:
Hi Duggy, just a humble comment as a jetson nano and raspberry pi user
We can build openvino binary for Jetson arm64 as Aznie mentions above but please note that anyway it cannot use Tegra gpu as an inference acceleration engine now and maybe in near(?) future.
It can run on cpu of which plugin has become available recently but performance is pretty poorer than vpu, or with vpu to add Intel NCS2 USB stick.
Some folks use Raspberry Pi with NCS2 for running OpenVINO in this kind of arm-based edge boxes since the runtime for raspbian arm32 and vpu has been provided by official site since early releases.
Just for your info:
Here is one example of FPS benchmark comparison for mobilenet-ssd model on openvino 2021.4, ubuntu 18.04.
jetson nano cpu : 5.1 fps
jetson nano vpu : 58 fps
raspberry pi 4B cpu : 4.2 fps
raspberry pi 4B vpu : 58 fps
i5-4670 WSL/Win10 cpu : 64 fps
Thank you for the feedback. I too have been playing with Raspberry Pi 4 and Nano Jetson Developer Kit 2GB. I was running OpenVino on my CPU machine but it was incredibly slow running multiple camera multiple targets (reidentification). Was hoping to get it running on the Nano Jetson to utilize the GPU processing for speed. Is this not a possibility? Alternatively was attempting same on the Raspberry PI but struggled to get that installed as well.
What would you recommend if I want to process multiple cameras for the multiple camera tracking with reidentification in a commercial environment and report the data back in real time? A Raspberry Pi 4 with NCS2 or Jetson Nano? Or alternatively just a pc with GPU?
There would be no available plugin on CUDA library as of this moment as far as I know.
If you want to evaluate on Jetson integrated gpu or nvidia discrete gpu board, OpenVINO SDK would not be your choice.
And if "CPU machine" means some recent generation Core i or Xeon processor in System Requirements list, I suppose it would be difficult to achieve better performance on Raspberry Pi with NCS2.
If you would like to evaluate your commercial application product with OpenVINO on various Intel hardware platforms, you may consider to try "Intel DevCloud for the Edge" program .
Thanks @okih, for sharing the information in the community. I believe it is useful for other community participants who are facing the same issue. Therefore, this thread will no longer be monitored since this issue has been resolved. If you need any additional information from Intel, please submit a new question.