OpenVINO Pre-trained Model Application Demos

#OpenVINO #Pre-trained Model #AI Application

In the previous “Introduction to OpenVINO”, I talked about what OpenVINO is, what it is capable of, how to get started, and performed integration with TensorFlow2. In today’s article, I would like to introduce some of the more practical applications that I built with the OpenVINO toolkit. All of the models that were used in this article could be downloaded from the OpenVINO Model Zoo or via the OpenVINO Model Downloader CLI.

Summary:

As promised in the previous article, over the two weeks, I came up with three applications and the source codes are now on Github. Nonetheless, the applications are not built for production environment, instead the code resembles as a starting point of major applications.

Before introducing the applications, here is how you would access and download the models using OpenVINO Toolkit. Link to OpenVINO Models Documentation.

Model Zoo - OpenVINO™ documentation

1. Initiate OpenVINO Environment
   <path to OpenVINO>\bin\setupvars.bat
2. Go to the path below:
   <path to OpenVINO>\deployment_tools\tools\model_downloader
3. In your command line interface:
   python downloader.py
   — — name “<model_name>”
   — — precisions “<model_precision>”
   — — output_dir “<path_to_your_desired_directory>”
4. Example:
   python downloader.py
   — name “person-vehicle-bike-detection-crossroad-yolov3–1020”
   — precisions “FP16”
   — output_dir “E:\TensorFlow\TensorFlow Projects\Project 005-TensorFlow with OpenVINO\model”

GitHub - CJsGit-tech/OpenVINO-Demo-2022-Applications

You can access all the source code from the link above. For the rest of the article, I will give introductions to all applications and their usage in the real world.

Car Counting Demo
Model(s):
person-vehicle-bike-detection-crossroad-1016
Description:
The goal of this application is to create an automated system that can perform real-time traffic monitoring. An extension of this application could be building a reinforcement learning AI that controls traffic lights. In other words, we can have AIs decide how the traffic flows which could improve overall traffic during rush hours.

Real-time Face Detection and Face Filter with StyleGAN
Model(s):
face-detection-retail-0005
emotions-recognition-retail-0003
style-GAN
Description:
The goal of this application is to demonstrate that OpenVINO has models that can perform
1) real-time face detection with masking using StyleGAN
2) emotion recognition.
These models are really common in our day-to-day lives. We see all smartphones now have face detection algorithms built into cameras. We see lots of different face filters in photoshop applications.

Text Detection and Eng Text Recognition
Model(s):
text-detection-0004
text-recognition-0014 (English Only)
Description:
The goal of this application is to demonstrate that OpenVINO has models that can perform Text Detection in images. There are also quite a few applications that are built upon the use of text detection models. For instance, inserting credit card information. In the past, manually typing the numbers and pins are necessary but more and more applications now allow the users to take a picture of the card! The information will be filled in automatically.

Side Note:
You will be able to find the text recognition part of the code in the Github repo.
Some other model demonstration results can be found in the YouTube Link Below.

Conclusion

I believe that OpenVINO toolkit is extremely powerful! Almost anyone will be able to build something out of the box. Overall, I will say that OpenVINO is a great tool to consider when Proof of Concept (PoC) is needed.

Developers can try out some pre-trained models and build a prototype in a short amount of time. The prototypes should help your team identify bottlenecks in later development stage.

Text Detection and Eng Text Recognition