Skip to content

README.md

Latest commit

 

History

History
62 lines (46 loc) · 6.72 KB

README.md

File metadata and controls

62 lines (46 loc) · 6.72 KB

Demo 01 - Licence plate recognition

Introduction

The first demo of this series is based on the "Red Hat OpenShift Data Science Workshop - Licence plate recognition"1 and shows the usage of Red Hat CodeReady Workspaces (CRW) running on Red Hat Openshift to create a development environment for an Artifical Intelligence (AI) workload, which can be executed from within CRW using CPU and GPU.

Requirements

In order to be able to create the CodeReady Workspace and run the sample AI workload as described below, the following requirements need to be fullfilled:

  • Red Hat OpenShift (tested on version 4.8.17)
  • Red Hat CodeReady Workspaces (tested on version 2.12.1)
  • Optional: NVIDIA GPU Operator (tested on version 1.8.2)

Usage

Creating the CodeReady workspace:

  1. Open CodeReady Workspaces (CRW) and click on "Create Workspace"

  2. Click "Custom Workspace" and in section "DevFile" paste the devfile URL for GPU or CPU execution in the field "Enter devfile URL":

    Then click "Load Devfile", adapt the settings in the yaml editor, if needed, and finally click "Create & Open". The first time you run the devfile it may take some minutes to create the workspace (e.g. due to downloading the proper container image).

  3. After the workspace has been created, you are asked if you trust the authors of the demo repo2. Click "Yes, I trust" and the demo repo's content is automatically being synced to your workspace (accessible on the left under "CHE (Workspace)").

Starting the webserver hosting the AI workload

  1. Go to the menu bar and select "Terminal" and "Run Task". Then click on "01_license_plate__Run server" at the top of the page. A command window is opening at the bottom of the page, where you can see the webserver for the AI workload being started. The first time you execute this task inside the workspace, the prerequistes are installed via Python's PyPi, which may take some minutes.
  2. After the AI webserver has been started successfully, a new window tells you that a new process is listening on port 5000, which is the desired AI webserver listener. You are asked if you want to create a redirect for this port, which you answer with "yes".
  3. Subsequently a new window pops up telling you that the redirect is enabled now and provides the URL to the webserver, which you can open by clicking "Open In New Tab". The new browser window should simply show the JSON string { "status": "ok"}, which means that the webserver and the AI workload are working properly. Please note that it may take some seconds until the redirect is working properly.

Testing the AI model

Go to the menu bar and select "Terminal" and "Run Task". Then click on "01_license_plate__Test AI model" at the top of the page. The number plate of picture "dev_demos/external/dataset/images/Cars374.png" is being analyzed by the AI model and the result should be the JSON String {'prediction': 'S7JDV'}. You can now play around with this AI model and test the reckognition of different number plates by opening the Python file "dev_demos/demos/01_license_plate/test_ai_model.py" in the CRW editor (under "CHE (WORKSPACE)" on the left side of the screen). Simply uncomment one of the examples provided and comment the other two. For example:

#response = requests.post('http://127.0.0.1:5000/predictions', '{"data": "Cars374.png"')
#response = requests.post('http://127.0.0.1:5000/predictions', '{"data": "Cars387.png"')
response = requests.post('http://127.0.0.1:5000/predictions', '{"data": "Cars18.png"}')

You can also add your own pictures to "dev_demos/external/dataset/images/" in the CRW editor and use them to test the AI model by replacing the string "Cars374.png" with the name of the newly uploaded file and executing the task "01_license_plate__Test AI model" as described above.

Testing the GPU computation

Go to the menu bar and select "Terminal" and "Run Task". Then click on "01_license_plate__Test GPU computation" at the top of the page. A new command window opens at the bottom of the screen, displaying the result of a simple matrix calculation as given in "dev_demos/demos/01_license_plate/test_gpu.py", which should be [[22. 28.][49. 64.]]. Additionally, the available GPUs are displayed. If [] is shown, no GPU could be found and the calculations have been done using the CPU.

You can play around with the file "test_gpu.py" and use with tf.device('/CPU:0'): or with tf.device('/GPU:0'): in front of the computations in order to run the workload with CPU or GPU, respectively. After running this file via task "01_license_plate__Test GPU computation" as described above, in the terminal output it will be clearly stated for each computation if a CPU or GPU has been used.

Note regarding external resources

The "Red Hat OpenShift Data Science Workshop - Licence plate recognition" repo3 has been included in this repo using git subtree as

git subtree add --prefix external/licence-plate-workshop https://github.com/rh-aiservices-bu/licence-plate-workshop.git main --squash

Starting from the original code's commit id 4ebb74326cdd70eedf861225f6604fef0d0e3194 changes regarding the PyPi packages in file external/licence-plate-workshop/requirements.txt have been done to fit to the environment at hand.

Note regarding secure endpoints

Please note that when creating endpoints on the fly using the IDE, i.e. a new window asking if you would like to create an endpoint for a new process listening on port 5000 (after having started the webserver as described above), then this endpoint is created by default with options public and not secure, meaning that everyone can access this endpoint if the exact URL is known (and access to the OpenShift cluster is available, of course). To create public endpoints which are protected, you can simply add an "endpoints" section to your component in the devfile with a secure: 'true' option as can be seen in the CPU-based devfile. After the workspace has been started (and the webserver is running) you can access the respective route by providing your workspace's JWT token, which can be found in the YAML representation of your workspace pod under "CHE_MACHINE_TOKEN". Using curl for example, calling the route is done by

curl -H "Authorization: bearer MY_JWT_TOKEN" https://routeXYZ-codeready.apps.MY_URL/

Footnotes

  1. https://rh-aiservices-bu.github.io/licence-plate-workshop/ ↩

  2. https://github.com/sa-mw-dach/dev_demos.git ↩

  3. https://github.com/rh-aiservices-bu/licence-plate-workshop.git ↩