opencv/doc/tutorials/dnn/dnn_openvino/dnn_openvino.markdown

OpenCV usage with OpenVINO

@prev_tutorial{tutorial_dnn_googlenet} @next_tutorial{tutorial_dnn_yolo}

Original author	Aleksandr Voron
Compatibility	OpenCV == 4.x

This tutorial provides OpenCV installation guidelines how to use OpenCV with OpenVINO.

Since 2021.1.1 release OpenVINO does not provide pre-built OpenCV. The change does not affect you if you are using OpenVINO runtime directly or OpenVINO samples: it does not have a strong dependency to OpenCV. However, if you are using Open Model Zoo demos or OpenVINO runtime as OpenCV DNN backend you need to get the OpenCV build.

There are 2 approaches how to get OpenCV:

• Install pre-built OpenCV from another sources: system repositories, pip, conda, homebrew. Generic pre-built OpenCV package may have several limitations:
  • OpenCV version may be out-of-date
  • OpenCV may not contain G-API module with enabled OpenVINO support (e.g. some OMZ demos use G-API functionality)
  • OpenCV may not be optimized for modern hardware (default builds need to cover wide range of hardware)
  • OpenCV may not support Intel TBB, Intel Media SDK
  • OpenCV DNN module may not use OpenVINO as an inference backend
• Build OpenCV from source code against specific version of OpenVINO. This approach solves the limitations mentioned above.

The instruction how to follow both approaches is provided in OpenCV wiki.