Getting openvino open source 2022/1.1 to work on Ubuntu 22.04

[rogerjames99]

I am trying to get openvino 2022/1.1 to work on Ubuntu 22.0 and talk to my NCS2 stick I have run into a problem. This is where I have got to so far.

1. Cloned the openvino repo.
2. Checked out the releases/2022/2.2 branch.
3. Initialised the submodules.
4. The build encountered a number compiler warnings that were treated as errors.
5. I fixed the error in src/core/src/type/bfloat16.cpp.
   `@@ -57,7 +57,7 @@ size_t bfloat16::size() const {
   }

bfloat16::operator float() const {

• uint32_t tmp = (static_cast<uint32_t>(m_value) << 16);

• static uint32_t tmp = (static_cast<uint32_t>(m_value) << 16);
  const float* f = reinterpret_cast<const float*>(&tmp);
  return *f;
  }`

6. I realised all the other errors were in a template in single include file and that it would not be easy to fix. So I turned off the treat warnings as errors flag in the CMake config.
7. I stuck my head in a bucket of cold water.
8. I got as far as building the python wheels and hit a problem with the numpy version. So I took a chance and modified src/bindings/python/requirements.txt to remove the top limit on the numpy version. This caused an error in src/bindings/python/wheel/CMakeLists.txt.

`@@ -33,7 +33,7 @@ endforeach()

execute_process(COMMAND ${PYTHON_EXECUTABLE} -c "import wheel.bdist_wheel ; print(f'{wheel.bdist_wheel.get_abi_tag()}')" OUTPUT_VARIABLE PYTHON_ABI)
execute_process(COMMAND ${PYTHON_EXECUTABLE} -c "import wheel.vendored.packaging.tags as tags ; print(f'{tags.interpreter_name()}{tags.interpreter_version()}')" OUTPUT_VARIABLE INTERPRETER)
-execute_process(COMMAND ${PYTHON_EXECUTABLE} -c "import wheel.vendored.packaging.tags as tags ; print(f'{next(tags._platform_tags())}')" OUTPUT_VARIABLE WHEEL_PLATFORM)
+execute_process(COMMAND ${PYTHON_EXECUTABLE} -c "import wheel.vendored.packaging.tags as tags ; print(f'{next(tags.platform_tags())}')" OUTPUT_VARIABLE WHEEL_PLATFORM)
string(STRIP ${PYTHON_ABI} PYTHON_ABI)
string(STRIP ${INTERPRETER} INTERPRETER)
string(STRIP ${WHEEL_PLATFORM} WHEEL_PLATFORM)
`

9. This finally got everything to build.
10. But as I expected there was a runtime error!

`Python 3.10.6 (main, May 29 2023, 11:10:38) [GCC 11.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.

from openvino.runtime import Core
/home/roger/.local/lib/python3.10/site-packages/openvino/runtime/utils/types.py:24: FutureWarning: In the future np.bool will be defined as the corresponding NumPy scalar.
(Type.boolean, np.bool),
Traceback (most recent call last):
File "", line 1, in
File "/home/roger/.local/lib/python3.10/site-packages/openvino/runtime/init.py", line 38, in
from openvino.runtime.ie_api import Core
File "/home/roger/.local/lib/python3.10/site-packages/openvino/runtime/ie_api.py", line 16, in
from openvino.runtime.utils.types import get_dtype
File "/home/roger/.local/lib/python3.10/site-packages/openvino/runtime/utils/types.py", line 24, in
(Type.boolean, np.bool),
File "/home/roger/.local/lib/python3.10/site-packages/numpy/init.py", line 313, in getattr
raise AttributeError(former_attrs[attr])
AttributeError: module 'numpy' has no attribute 'bool'.
np.bool was a deprecated alias for the builtin bool. To avoid this error in existing code, use bool by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use np.bool_ here.
The aliases was originally deprecated in NumPy 1.20; for more details and guidance see the original release note at:
https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations. Did you mean: 'bool_'?

`

Numpy strikes again.

Has anyone been down this path before?
Has anyone got any ideas on how to get past with minimal intervention in the source?

Help!

Roger

[berrylizzard]

Maybe this will help. I've gotten OpenVINO 2022.2 working on Ubuntu 22.04. Following (somewhat) the guide from here to install runtime, then pip install openvino-dev[tensorflow2]==2022.2.0 for the dev toolkit. Note, that 2022.2.0 supports Python 3.6-3.9 so I'm using Python 3.9.16. I also made sure to have numpy (v1.23.1) installed before these steps.
Here's how:

1. Download file l_openvino_toolkit_ubuntu20_2022.2.0.7713.af16ea1d79a_x86_64.tgz (94.6 MB) from archive and rename it for ease of use to openvino_2022.2.0.tgz
2. Unzip it and move to a known directory (I created an ~/intel/ folder)

$ tar -xf openvino_2022.2.0.tgz
$ mv l_openvino_toolkit_ubuntu20_2022.2.0.7713.af16ea1d79a_x86_64 ~/intel/openvino_2022.2.0
$ cd ~/intel/openvino_2022.2.0/install_dependencies

3. Download the install_openvino_dependencies.sh file from here.
   Rename it to install_openvino_dependencies_2022.3.0.sh and move it to ~/intel/openvino_2022.2.0./install_dependencies/ (to avoid “Unsupported OS: ubuntu22.04” error; 2022.3.0 does support Ubuntu 22.04).

$ chmod +x install_openvino_dependencies_2022.3.0.sh
$ sudo -E ./install_openvino_dependencies_2022.3.0.sh
$ cd 
$ source intel/openvino_2022.2.0/setupvars.sh

5. Check:

$ python -c "from openvino import runtime; print(runtime.get_version())"
2022.2.0-7713-af16ea1d79a-releases/2022/2

6. Now install dev toolkit (reference) with pip install openvino-dev[tensorflow2]==2022.2.0 and check that it works with mo -h.

[rogerjames99]

Thanks berrylizard,

Your info will be really useful. I had literally just started down the same path by cloning a fresh copy of the repo and checking out the releases/2202/3 branch. I have just checked and the build of openvino has finished succesfully. I will now follow your instructions for the dev toolkit.

Roger

[rogerjames99]

I got a succesful build with the following.

git clone https://github.com/openvinotoolkit/openvino.git myvino2
cd myvino2
git checkout releases/2022/3
mkdir build
python -m venv build/vino
build/vino/bin/activate
git submodule update --init --recursive
cd build
ccmake -DCMAKE_BUILD_TYPE=Debug ..
make -j$(nproc --all)
pip install openvino-dev[tensorflow2]==2022.3

But it is not seeing my NCS2 stick.

(vino) roger@dragon:~/myvino2/build$ python
Python 3.10.6 (main, May 29 2023, 11:10:38) [GCC 11.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.

import openvino.runtime as ov
core = ov.Core()
print (core.available_devices)
['CPU']

I set the log level to trace.

export OPENVINO_LOG_LEVEL=5

But no logging was visible. I have not got a clue how this logging system is supposed to work or where the output goes to. Any help on this would be gratefully received.

As far as I know the myriad udev rules are set up correctly. The device is visible in lsusb. It also works as expected when mapped into an Ubuntu 20.04 vm running the 2022/1.1 release.

Roger