OpenVINO

Set up a conda/venv virtual env and install dependencies from requirements.txt

OpenVINO install and inference

OpenVINO PyPi Package

Install with pip using: pip install openvino from https://pypi.org/project/openvino/

OpenVINO Python Inference Samples

https://github.com/odundar/openvino_python_samples

Running sample python scripts

1. Person/Object Detection

# image mode
$ python openvino_person_detection.py -m image -i PATH_TO_IMG -o OUTPUT_DIR
# video mode
$ python openvino_person_detection.py -m video -i PATH_TO_VID -o OUTPUT_DIR
# webcam mode
$ python openvino_person_detection.py -m webcam -o OUTPUT_DIR

2. Person Re-identification

# video mode
$ python openvino_person_reidentification.py -m video -i PATH_TO_VID -o OUTPUT_DIR

3. Person Action Counting (Punch/Kick/Normal)

# video mode
$ python openvino_person_action_counting.py -m video -i PATH_TO_VID -o OUTPUT_DIR

Pretrained OpenVINO models

1. OpenVINO Model Documentation and Download

https://docs.openvinotoolkit.org/2021.3/omz_models_group_intel.html

2. OpenVINO GitHub Public Model Repo

https://github.com/openvinotoolkit/open_model_zoo/tree/master/models/public

3. Download Intel OpenVINO IR (Intermediate Representation) models

• OpenVino Workbench with Docker (Recommended) https://docs.openvino.ai/latest/workbench_docs_Workbench_DG_Introduction.html

• Python openvino-dev pip pkg https://github.com/openvinotoolkit/open_model_zoo/blob/master/tools/model_tools/README.md

Creating OpenVINO models from other NN frameworks

1. Convert to OpenVINO format with Docker

Local repository available in ./google_teachable_machine_to_openvino

https://github.com/ojjsaw/teachable-machine-openvino.git

2. OpenVINO training extensions

https://github.com/openvinotoolkit/training_extensions

3. Deep Learning Workbench from DockerHub

Download & Convert to IR models for OpenVINO Inference

https://docs.openvinotoolkit.org/latest/workbench_docs_Workbench_DG_Install_from_Docker_Hub.html