workshop-py-face

Example about an IRIS production using Embedded Python to recognize and identify faces in photos

You can find more in-depth information in https://learning.intersystems.com.

What do you need to install?

• Git
• Docker (if you are using Windows, make sure you set your Docker installation to use “Linux containers”).
• Docker Compose
• Visual Studio Code + InterSystems ObjectScript VSCode Extension

Setup

Build the image we will use during the workshop:

$ git clone https://github.com/intersystems-ib/workshop-py-face
$ cd workshop-py-face
$ docker-compose build

The current project is self-deployable and it doesn’t require further configuration. The Python libraries required are installed during the deployment of the container.

Example

The main purpose of this example is to ingest images from a specific folder, identify the faces in the images and check any possible match of the faces in a group of images used as repository.

In order to accomplish this goal we are going to use the functionality available in IRIS to use a language like Python in our ObjectScript classes.

We are going to use two models pre-trained to simplify the code, the first one, mobilenet_graph.pb will help us to recognize all the faces in our image. The second one, facenet_keras_weights.h5 will allow us to compare the face found with the faces in the images of our repository.

Nomenclator Production

• Run the containers we will use in the workshop:

docker-compose up

Automatically an IRIS instance will be deployed and a production will be configured and run to read JPG files from /shared/JPG folder.

• Open the Management Portal.
• Login using the default superuser/ SYS account.
• Click on Nomenclator Production to access the sample production that we are going to use. You can access also from NOMENCLATOR namespace through Interoperability > User > Configure > Production.
• Click on Nomenclator.FaceChecker business service and review the configuration.
  • You will notice that is using EnsLib.File.InboundAdapter as the adapter to read all the files in /shared/JPG folder.
  • Also you will see five new settings:
    • KnownsPath: the folder that will be used as a repository of knowns faces by the Python code. The Python method will compare the found image with the images in this folder.
    • UnknownsPath: any file found in /shared/JPG will be copied to this folder to be treated by the Python method.
    • ResultsPath: in this folder we are going to create a copy of the original image with a mark around any face found out. In case of match with any other face the mark will be of green color, otherwise the color will be red.
    • ModelDetectorFilePath: location in the container of mobilenet_graph.pb file.
    • ModelRecognitionFilePath: location in the container of facenet_keras_weights.h5 file.

Opening Visual Studio Code you will be able to manage the different folders configured, you will be able to copy jpg images from your own computer into the container folder /shared/knowns for all the knowns pictures and /shared/JPG for all the pictures that we want identify.

Python Method

If you are interested in know more about the Python functionality you can take a look to this article in spanish that I used to build the entire functionality.

How to make it run

You only need to copy the file travi.jpg located in /shared/test folder into /shared/JPG folder. I’ve included 4 photos of John Travolta as the face known, so any new photo added into /shared/JPG of John Travolta should return in the Events Log a message with the match and you will see in /shared/results folder your photo with the face enmarked in green.

In the case that you put into /shared/JPG the other file nicolas.jpg you will see an empty line in the Events Log and the JPG created into /shared/results will contain the face enmarked in red.