[Q&A] OpenVINO with Jupyter Notebooks

[Helio_Guerra]

Q: Could you share the notebooks?
A: The notebooks are here: https://github.com/openvinotoolkit/openvino_notebooks 

Q: Is this the OpenCV which comes with OpenVINO?
A: OpenCV is a requirement of openvino-dev so it will be installed when you do pip install openvino-dev, but the version may be different than the version that is included in the OpenVINO installer

Q: Can OpenVINO also be used for audio analysis?
A: Yes, it can be. Please take a look at the speech demo - https://github.com/openvinotoolkit/open_model_zoo/tree/master/demos/speech_recognition_demo/python 

Q: Which version of OpenVINO will run if I follow webinar instructions? 
A: The instructions install the latest version of OpenVINO (currently 2021.3), and the notebooks are validated to work with that version.

Q: How do you track how this python package relate to C++ OpenVINO releases? Are they 1-1 connection?
A: Python packages include all required C++ libraries, and C++ OpenVINO releases are aligned with PyPI releases. We use the released versions (the latest available) only to configure notebooks, so the connection is 1-1.

Q: What OS is supported on Raspberry Pi by OpenVINO?
A: Debian 9

Q:  What is the basic language knowledge requirement for OpenVINO?
A: OpenVINO provides C++/C and Python API

Q: Can the notebook utilize GPU or just CPU?
A: The notebooks can definitely use the GPU. Most notebooks let you specify CPU or GPU in the settings, so running on CPU is changing three letters. 

Q: Can the monodepth be used to extract an object via masking?
A: You can, but it depends on the image how well that works. The U2Net model demoed in webinar is much better for this.

Q: Does U2Net work for object segmentation (e.g. objects in the foreground)?
A: Yes, the current U2Net notebook supports salient object segmentation and human segmentation (it is a toggle in the notebook). By default the notebook uses the lite salient model, which is smaller, but less accurate. Please try the full U2Net model if this is something you're interested in.

Q: Where do we find the U2Net for OpenVINO?
A: It is here: https://github.com/openvinotoolkit/openvino_notebooks/tree/main/notebooks/205-vision-background-removal It was added on April 22, 2021, so if you happened to clone our repository before today, please do a git pull and install the requirements again.

Q: Can "step 2" of demo be done in Anaconda environment?
A: Conda distro doesn't include the package with OpenVINO tools (Model Optimizer, Post-training Optimization Tool, Accuracy Checker), but you can install OpenVINO using pip, but Conda installation doesn't include python3-dev package and you would need to install it manually (yum/apt). We are recommending to use python installations.

Q: Could you update the notebooks for all examples from OpenVINO package?
A: Thanks for your interest in the notebooks! We actually have five now, as you just saw, and we definitely do not plan to stop here. We will add many more notebooks. Please, let us know which models you would like to see most.

Q: What hardware can i use to get started with this? 
A: You can find system requirements here: https://software.intel.com/content/www/us/en/develop/tools/openvino-toolkit/system-requirements.html 

Q: You don't need to do that download in order to run the notebooks, correct?
A: That is correct! You only need the openvino-dev pip package, not the full Windows installer. All instructions are in the README of our repository.

Q: Would you be able to run it on Macs with new M1 chip?
A: We do not officially support this, but some notebooks are known to work with a workaround, and we will keep testing this unofficially. See https://github.com/openvinotoolkit/openvino_notebooks/discussions/10 

Q: What solutions would you recommend for developers that need to create OpenVINO SaaS solutions?
A: Check this out https://github.com/openvinotoolkit/model_server