tearaikazuki/ovaas-backend

OVaaS Back-end, powered by Azure functions.

OVaaS Backend Development

This is a OVaaS's backend source code. Please follow the instrunctions described below and try to have your custom backend application. By the way, the backend of OVaaS consists of Azure Functions and Azure VM ScaleSet. And other than them, various technologies, softwares and services are included. You need to understand the overview of them as a first step.

0. Prerequisites

• Windows 10 / Linux / macOS
• Python 3.7+
• Visual Studio Code
• Docker
• Node.js (to use npm)
• Azure Storage Explorer
• Azure Functions Core Tools (v3.x)

1. Setup Development Environment

Install softwares described as prerequisites

Recommend to install latest version of each software. All softwares can be installed easily with GUI-based install wizard.

Install VSCode Extensions

• Launch VSCode
• Go to the Extention pane
• Search and install the extentions below
  • Azure Functions
  • REST Client

Clone this repository and open it with VSCode

• Launch command prompt and change directory as you want.
• Clone this repository there by the command below

  git clone https://github.com/OVaaS/ovaas-backend.git

• Launch VSCode to load the repository by the command below

  cd ovaas-backend
  code .

Create Python venv on VSCode

VSCode shows a dialog to ask you if venv can be automatically created so it will be easy to create it by clicking "Yes" button there. But you can create it by yourselves using the instructions below if it is not shown.

• Launch command prompt in VSCode
• Create venv by the command below This command will create a venv named ".venv"

  python -m venv .venv

After creating the venv, you need to choose a python interpreter to run an application on VSCode.

• Choose Python interpreter

  Open the "Command Palette" from the "View" item on the menu bar and type "Python: Select Interpreter", then you can choose a python interpreter from the list shown there. Recommend to choose the one in the venv you just created.

• Install required libraries

  python -m pip install --upgrade pip
  pip install -r requirements.txt
  pip install azure-storage-blob

Launch a local Azure Storage as a Docker container

Now you are ready to launch some docker containers for developping. The first one is a local Azure Storage. Do it following the command below.

• Launch a local Azure storage using the Azurite Docker image

  Windows 10

  scripts\LaunchAzurite.bat
  

  Linux / macOS

  sh scripts/LaunchAzurite.sh

  After the command executed, check if local Azure storage is launched by Azure Storage Explorer.

Now you are ready for running a sample application. Follow the next step.

2. Run the sample applications

Launch a local OpenVINO Model server with a pre-trained model

• Look at the list of all pre-trained models by the command below.

  Windows 10

  scripts\GetModelList.bat 1st-parameter
  
  rem for latest OpenVINO model zoo
  scripts\GetModelList.bat latest
  rem for ver 2021.1 OpenVINO model zoo
  scripts\GetModelList.bat 2021.1
  

  Linux / macOS

  scripts/GetModelList.sh 1st-parameter
  
  # for latest OpenVINO model zoo
  scripts/GetModelList.sh latest
  # for ver 2021.1 OpenVINO model zoo
  scripts/GetModelList.sh 2021.1

  Here you need one parameter.

  • --1st parameter: The version of OpenVINO. Mostly "latest" works fine but sometimes specific model is provided only in specific version of OpenVINO. For example, our "Colorization" model is provided in the version "2021.1". That's why above example shows "2021.1" in command line.

• Choose one model to download and download it by the command below. For example, current applications use these models.

  Windows 10

  scripts\DownloadModel.bat 1st-parameter 2nd-parameter
  
  rem For Humanpose
  scripts\DownloadModel.bat human-pose-estimation-0001 latest
  rem For Handwritten
  scripts\DownloadModel.bat handwritten-japanese-recognition-0001 latest
  rem For Colorization
  scripts\DownloadModel.bat colorization-v2 2021.1
  rem For Object Detection
  rem Object detection is original model not in Open Model Zoo.
  

  Linux / macOS

  scripts/DownloadModel.sh 1st-parameter 2nd-parameter
  
  # For Humanpose
  scripts/DownloadModel.sh human-pose-estimation-0001 latest
  # For Handwritten
  scripts/DownloadModel.sh handwritten-japanese-recognition-0001 latest
  # For Colorization
  scripts/DownloadModel.sh colorization-v2 2021.1
  # For Object Detection
  # Object detection is original model not in Open Model Zoo.

  Here you need two parameters.

  • 1st-parameter: The model name
  • 2nd-parameter: The version of OpenVINO. Mostly "latest" works fine but sometimes specific model is provided in only specific version of OpenVINO. For example, our "Colorization" model is provided in the version "2021.1". That's why above example shows "2021.1" in command line.

• Check if the model is downloaded in local folder. For example, above "human-pose-estimation-0001" should be here.

  PARENT_DIR/ovaas-backend/models/intel/human-pose-estimation-0001/FPXX

• Copy the absolute path to the XML file and the BIN file of the pre-trained model and upload the model to the local Azure Storage by the command below.

  Windows

  python scripts\UploadModelFilesToAzureStorage.py --model_name human-pose-estimation --xml_file_path PARENT_DIR\ovaas-backend\models\intel\human-pose-estimation-0001\FPXX\human-pose-estimation-0001.xml --bin_file_path PARENT_DIR\ovaas-backend\models\intel\human-pose-estimation-0001\FPXX\human-pose-estimation-0001.bin
  

  Linux / macOS

  python3 scripts/UploadModelFilesToAzureStorage.py --model_name human-pose-estimation --xml_file_path PARENT_DIR/ovaas-backend/models/intel/human-pose-estimation-0001/FPXX/human-pose-estimation-0001.xml --bin_file_path PARENT_DIR/ovaas-backend/models/intel/human-pose-estimation-0001/FPXX/human-pose-estimation-0001.bin

  Here you need four parameters.

  • --model_name: The unique model name
  • --xml_file_path: The absolute path to the XML file of the pre-trained model
  • --bin_file_path: The absolute path to the BIN file of the pre-trained model
  • --connection_string: Optional. The connection string to access the local Azure Storage. You can get this on Azure Storage explorer but probably don't need to specify it. You can edit source code if any other connection string is needed.

• Launch a local OpenVINO model server

  Windows 10

  scripts\LaunchOVMS.bat 1st-parameter 2nd-parameter 3rd-parameter 4th-parameter
  
  rem Example: Humanpose
  scripts\LaunchOVMS.bat human-pose-estimation 192.168.10.107 9000 latest
  
  rem Example: Colorization
  scripts\LaunchOVMS.bat colorization 192.168.10.107 9000 2021.1
  

  Linux / macOS

  sh scripts/LaunchOVMS.sh 1st-parameter 2nd-parameter 3rd-parameter 4th-parameter
  
  # Example: Humanpose
  sh scripts/LaunchOVMS.sh human-pose-estimation 192.168.10.107 9000 latest
  
  # Example: Colorization
  sh scripts/LaunchOVMS.sh colorization 192.168.10.107 9000 2021.1

  Here you need four parameters.

  • 1st-parameter: The unique model name you just named when to upload the model to the local Azure storage.
  • 2nd-parameter: The IP address assigned to your PC's ethernet adapter. Note: "localhost" and "127.0.0.1" will not work fine.
  • 3rd-parameter: The port number to communicate to a model server.
  • 4th-parameter: The version of OpenVINO. Mostly "latest" works fine but sometimes specific model is provided in only specific version of OpenVINO. Set same version as you download the model in previous step.

Launch an Azure functions emulater on VSCode

• From the "Run" on the menu bar, click "Start Debugging". Then the emulater should start automatically. You will see the logs like below if it starts successfully.

  Azure Functions Core Tools
  Core Tools Version:       3.0.3442 Commit hash: 6bfab24b2743f8421475d996402c398d2fe4a9e0  (64-bit)
  Function Runtime Version: 3.0.15417.0
  
  
  Functions:
  
          HttpTriggerHumanPose: [GET,POST] http://localhost:7071/api/HttpTriggerHumanPose
  
  For detailed output, run func with --verbose flag.
  [2021-05-10T02:00:03.532Z] Worker process started and initialized.
  [2021-05-10T02:00:03.721Z] Host lock lease acquired by instance ID '000000000000000000000000F6FB3AFD'.

Run the sample application

• Open the file called "request.http".
• Click the "Send Request" on the top of the pane, then you can see the inference result after a few seconds on the other pane opened like the image below.

3. Develop your custom application

Follow the Step 2 to launch a new OpenVINO model server

You need to launch a new OpenVINO model server with your desired pre-trained model. To do that, follow the step 2 again and have a model server started. Note that you need to use different port number from the one used on human-pose-estiimation model server.

Create a new Azure Function in the project

• Click the Azure button on the bar on the left of VSCode
• Click the "Create Function" button, then you need to input information below.
  1. Choose "HTTP trigger"
  2. Type an unique function name
  3. Choose "Anounymous" as Authorization Level
• Then you will have a new function template in the same project.

Write a new application code following sample code

Basically you can imitate the sample human-pose-estimation's init.py.

Add IP and PORT setting to local.settings.json

If you would like to add new backend aapplication called "AI App", then add new IP address and port number setting into the file "local.settings.json" as below. Actually IP address should be localhost everytime for local development and port number should be different from the numbers already assigned. This example shows 9004 is assigned. These variables will be refered in a source code you will made in init.py.

{
  "IsEncrypted": false,
  "Values": {
    "FUNCTIONS_WORKER_RUNTIME": "python",
    "HUMANPOSE_IPADDRESS": "localhost",
    "HUMANPOSE_PORT": "9000",
    "HANDWRITTEN_IPADDRESS": "localhost",
    "HANDWRITTEN_PORT": "9001",
    "COLORIZATION_IPADDRESS": "localhost",
    "COLORIZATION_PORT": "9002",
    "OBJECTDETECTION_IPADDRESS": "localhost",
    "OBJECTDETECTION_PORT": "9003",
    "AIAPP_IPADDRESS": "localhost",
    "AIAPP_PORT": "9004"
  }
}

Launch Azure Function Emulator and test it from request.http

This is same as wht you have done on running the sample above. This would be not difficult. Enjoy anyway!