Azure functions for the configurable image recognition and classification pipeline

Details on how to develop for Azure function and the Azure Function tools for Visual Studio

Each function is added to its own folder and contains the following artifacts:

function.json   // your function binding
project.json    // your project dependencies
run.csx         // your code
test.json       // your test message

Here an example implementation for samplefunction:

#load "..\Common\FunctionHelper.csx"

using System;
using System.Configuration;
using Newtonsoft.Json;
using Newtonsoft.Json.Linq;
using Microsoft.WindowsAzure.Storage;
using Microsoft.WindowsAzure.Storage.Queue;

// your classifier's name
private const string ClassifierName = "samplefunction";

public static void Run(string inputMsg, TraceWriter log)
{
    log.Verbose($"Process {inputMsg}");
    PipelineHelper.Process(SampleFunction, ClassifierName, inputMsg, log);
}
public static dynamic SampleFunction(dynamic inputJson, string imageUrl, TraceWriter log)
{ 
    // TODO: do your processing here and return the results

    return new {
        stringOutput = "your output string value",
        intOutput = 10 
    };
}

As the last step, you need to declare the functions dependencies in dependencies.json. In the example below, facematch depends on faceprint and faceprint depends on facedetection.

{
  "facecrop": ["facedetection"],
  "facedetection": [],
  "facematch": ["faceprint"],
  "faceprint": ["facedetection"],
  "generalclassification": [],
  "ocr": []
}

Therefore the pipeline definition for facematch would look like "facedetection,faceprint,facematch"

Each function should contain a simple test message stored in test.json:

{
  "job_definition": {
    "batch_id": "mybatchid",
    "id": "myjobid",
    "input": {
      "image_url": "your image url",
      "image_classifiers": [ "classifier1", "classifier2" ]
    },
    "processing_pipeline": [ "your_input_queue_name", "your_output_queue_name" ],
    "processing_step": 0

  },
  "job_output" : {
    "sample_job1" : {
      "output": "sample_job1 output"
    }
  }
}

Ensure you add all needed input data (your dependencies) as part of the job_output and set your_input_queue_name and your_input_queue_name for the processing_pipeline property.

To test your functions locally, you need to update your appsettings.json file:

  {
    "IsEncrypted": false,
    "Values": {
      "FACES_CONTAINER": "faces",
      "SQL_CONNECTION_STRING": "<YOUR_SQL_CONNECTION_STRING>",
      "FACE_API_KEY": "<YOUR_FACE_API_KEY>",
      "VISION_API_KEY": "<YOUR_VISION_API_KEY",
      "AzureWebJobsStorage": "<YOUR_STORAGE_CONNECTION_STRING>",
      "AzureWebJobsDashboard": "<YOUR_STORAGE_CONNECTION_STRING>"
    }
  }

Once you've done that, you can either run your functions in Visual Studio or use azure-functions-cli directly:

npm i -g azure-functions-cli

To run your function, just execute

func run YourFunctionName

to run the tests, run the Imaginem-Cli. The following triggers the SampleFunction by adding its test.json to the sample queue.

.\Imaginem-Cli.exe test SampleFunction sample