The current version of OpenVINO™ Training Extensions was tested in the following environment:

  • Ubuntu 20.04

  • Python 3.8 ~ 3.10

  • (Optional) To use the NVidia GPU for the training: CUDA Toolkit 11.7


If using CUDA, make sure you are using a proper driver version. To do so, use ls -la /usr/local | grep cuda.

If necessary, install CUDA 11.7 (requires ‘sudo’ permission) and select it with export CUDA_HOME=/usr/local/cuda-11.7.

Install OpenVINO™ Training Extensions for users#

1. Clone the training_extensions repository with the following command:

git clone
cd training_extensions
git checkout develop

2. Set up a virtual environment.

# Create virtual env.
python -m venv .otx

# Activate virtual env.
source .otx/bin/activate

3. Install PyTorch according to your system environment. Refer to the official installation guide


Currently, only torch==1.13.1 ~ 2.0.1 was fully validated. (older versions are not supported due to security issues).

# Install command for torch==2.0.1 for CUDA 11.7:
pip install torch==2.0.1 torchvision==0.15.2 --extra-index-url

# Or, install command for torch==1.13.1 for CUDA 11.7:
pip install torch==1.13.1 torchvision==0.14.1 --extra-index-url
  1. Install OpenVINO™ Training Extensions package from either:

  • A local source in development mode

pip install -e .[full]
  • PyPI

pip install otx[full]

5. Once the package is installed in the virtual environment, you can use full OpenVINO™ Training Extensions command line functionality.

Install OpenVINO™ Training Extensions for developers#

Install tox and create a development environment:

pip install tox
# -- need to replace '310' below if another python version needed
tox devenv venv/otx -e tests-all-py310
source venv/otx/bin/activate

Then you may change code, and all fixes will be directly applied to the editable package.

Install OpenVINO™ Training Extensions by using Docker#

To build a docker image with Python 3.9, run a command below from the working copy of the OpenVINO training extensions.

# build a docker image (otx/cpu/python3.9:latest) with Python 3.9 (default)
training_extensions$ ./docker/
# or, with other version of Python e.g., 3.10
training_extensions$ ./docker/ --python 3.10


When the docker image build script completed successfully, the image will be named and tagged as otx/cpu/python<py-version-string>:latest. You can check it using the command docker images on the terminal.

To start the OpenVINO training extensions container using the image built in above, run a command below.

# start a container from `otx/cpu/python3.9:latest' image.
$ docker run \
    -it \ # enter interactive terminal
    --rm \ # remove container after use
    -v "$(pwd):/mnt/shared:rw" \ # mount current folder on host machine to the container
    --shm-size=4g \ # increase mounted shared memory
    otx/cpu/python3.9:latest    # name of the docker image to be used to create container

Enjoy OpenVINO training extensions!

# find all templates for the classification task
root@fc01132c3753:/training_extensions# otx find --task classification
|      TASK      |                         ID                        |          NAME         |                                       BASE PATH                                       |
| CLASSIFICATION |       Custom_Image_Classification_DeiT-Tiny       |       DeiT-Tiny       |           src/otx/algorithms/classification/configs/deit_tiny/template.yaml           |
| CLASSIFICATION |    Custom_Image_Classification_EfficinetNet-B0    |    EfficientNet-B0    |    src/otx/algorithms/classification/configs/efficientnet_b0_cls_incr/template.yaml   |
| CLASSIFICATION |   Custom_Image_Classification_EfficientNet-V2-S   |   EfficientNet-V2-S   |   src/otx/algorithms/classification/configs/efficientnet_v2_s_cls_incr/template.yaml  |
| CLASSIFICATION | Custom_Image_Classification_MobileNet-V3-large-1x | MobileNet-V3-large-1x | src/otx/algorithms/classification/configs/mobilenet_v3_large_1_cls_incr/template.yaml |

Run tests#

To run some tests, need to have development environment on your host. The development requirements file (requirements/dev.txt) would be used to setup them.

$ pip install -r requirements/dev.txt
$ pytest tests/

Another option to run the tests is using the testing automation tool tox. Following commands will install the tool tox to your host and run all test codes inside of tests/ folder.

$ pip install tox
$ tox -e tests-all-py310-pt1 -- tests/


When running the tox command above first time, it will create virtual env by installing all dependencies of this project into the newly created environment for your testing before running the actual testing. So, it is expected to wait more than 10 minutes before to see the actual testing results.


1. If you have problems when you try to use pip install command, please update pip version by following command:

python -m pip install --upgrade pip

2. If you’re facing a problem with torch or mmcv installation, please check that your CUDA version is compatible with torch version. Consider updating CUDA and CUDA drivers if needed. Check the command example to install CUDA 11.7 with drivers on Ubuntu 20.04.

3. If you use Anaconda environment, you should consider that OpenVINO has limited Conda support for Python 3.6 and 3.7 versions only. So to use these python versions, please use other tools to create the environment (like venv or virtualenv) and use pip as a package manager.

4. If you have access to the Internet through the proxy server only, please use pip with proxy call as demonstrated by command below:

python -m pip install --proxy http://<usr_name>:<password>@<proxyserver_name>:<port#> <pkg_name>