Implement a pipeline¶

This tutorial shows how to implement a Pipeline in the Swiss AI Center project step by step. It will guide you through the process of creating a Pipeline to detect faces in an image and blur them.

Tutorial¶

Prerequisites¶

To follow this tutorial, you need to have the following tools installed:

• Python 3.10
• An IDE (e.g. Visual Studio Code)

Prepare the Core Engine¶

To implement a Pipeline, you need to have a running Core Engine. If you don't have one, you can follow the Core Engine documentation.

Prepare the services¶

The first step is to prepare the services that will be used in the Pipeline. In this tutorial, we will use the following services:

• Face detection
• Image blur

Launch the services¶

If the services are not running you can follow the explanations in the reference documentation to start them.

• Face detection
• Image blur

Create the pipeline¶

The Pipeline is created by posting a JSON object to the /pipelines endpoint of the Core Engine. Create a file named face-blur-pipeline.json in your IDE with the following code:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39

{
    "name": "Face Blur",
    "slug": "face-blur",
    "summary": "Face Blur",
    "description": "Face Blur",
    "data_in_fields": [
        {
            "name": "image",
            "type": [
                "image/jpeg",
                "image/png"
            ]
        }
    ],
    "data_out_fields": [
        {
            "name": "result",
            "type": [
                "image/jpeg",
                "image/png"
            ]
        }
    ],
    "steps": [
        {
            "identifier": "face-detection",
            "needs": [],
            "inputs": ["pipeline.image"],
            "service_slug": "face-detection" // change this with the slug of your face detection service
        },
        {
            "identifier": "image-blur",
            "needs": ["face-detection"],
            "condition": "len(face-detection.result['areas']) > 0",
            "inputs": ["pipeline.image", "face-detection.result"],
            "service_slug": "image-blur" // change this with the slug of your image blur service
        }
    ]
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14

    [
        {
            "created_at": "2023-06-01T13:55:15.936033",
            "updated_at": "2023-06-01T13:55:19.831817",
            "name": "Face Detection",
            "slug": "face-detection",
        },
        {
            "created_at": "2023-06-01T13:55:15.936033",
            "updated_at": "2023-06-01T13:55:19.800560",
            "name": "Image Blur",
            "slug": "image-blur",
        }
    ]

What we just did is to create a Pipeline with two steps. The first step is the face detection service and the second step is the image blur service. The second step will only be executed if the first step detects at least one face. The Pipeline will take an image as input and return an image as output.

The inputs of each step are the outputs of the previous steps. The first step takes the pipeline's image as input and the second step takes the Pipeline's image and the result of the face detection as input.

Post the pipeline¶

Now that we have our Pipeline, we can post it to the Core Engine. To do so, go to the FastAPI documentation of the running Core Engine and use the /pipelines endpoint to post the Pipeline by clicking on the Try it out button

Simply copy the content of the face-blur-pipeline.json file and paste it in the body field of the /pipelines endpoint and click on the Execute button.

You should receive a 200 response with the Pipelineine you just posted.

Run the pipeline¶

You can run the pipeline using the FastAPI Swagger interface or by using the Webapp.

Using the FastAPI Swagger interface¶

Now that we have our Pipeline, we can run it. To do so, go to the FastAPI documentation of the running Core Engine and you should see the Pipeline you just posted in the Registered pipelines endpoint with the slug /face-blur.

Click on the Try it out button, add an image to the body and click on the Execute button.

You should receive a 200 response with a Pipeline Execution object in the response body. This object contains the id of the execution and the tasks that will be executed.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

{
  "pipeline_id": "2175ae79-4e48-4d1b-97df-0bcbba4c5d2b",
  "current_pipeline_step_id": "e0028cf9-0b62-48b4-b0c7-b91ec930d083",
  ...
  "tasks": [
    {
      "data_in": [
        "e930424b-63a8-4a4e-b49c-54eb50cd1996.jpg"
      ],
      "data_out": null,
      "status": "pending",
      "service_id": "35ace881-6673-4fb6-b454-2f94b1547fd6",
      "pipeline_execution_id": "a17dd3ef-0682-4154-baaa-cb524650f6f4",
      "id": "01636a67-b78e-41a6-9127-9efa1f0c1a9a"
    },
    {
      "data_in": null,
      "data_out": null,
      "status": "scheduled",
      "service_id": "6a20b1b7-ef3d-4a01-bf05-409558bda916",
      "pipeline_execution_id": "a17dd3ef-0682-4154-baaa-cb524650f6f4",
      "id": "9557c201-477f-45c6-8b8a-93ce93079f74"
    }    
  ]
}

You can check the status of the execution by checking the status of the last task with the /tasks/{task_id} endpoint. You can find the id of the last task in the tasks array of the Pipeline execution object.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

 {
  "created_at": "2023-06-06T14:45:25.517644",
  "updated_at": "2023-06-13T11:50:10.151580",
  "data_in": [
    "e930424b-63a8-4a4e-b49c-54eb50cd1996.jpg",
    "2594162e-7927-4dc9-9a22-d135154f9e93.json"
  ],
  "data_out": [
    "bfc86360-ad3f-4249-aaa1-21d36e84f612.jpg"
  ],
  "status": "finished",
  "service_id": "6a20b1b7-ef3d-4a01-bf05-409558bda916",
  "pipeline_execution_id": "a17dd3ef-0682-4154-baaa-cb524650f6f4",
  "id": "9557c201-477f-45c6-8b8a-93ce93079f74",
  ...
  "pipeline_execution": {
    "created_at": "2023-06-06T14:45:25.517644",
    "updated_at": "2023-06-13T11:50:10.156728",
    "pipeline_id": "2175ae79-4e48-4d1b-97df-0bcbba4c5d2b",
    "current_pipeline_step_id": null,
    "id": "a17dd3ef-0682-4154-baaa-cb524650f6f4",
    "files": null
  }
}

Once the status of the last task is finished, you can download the result by copying the data_out file keys and use them with the /storage/{key} endpoint.

If the picture you provided had a face, the result should be blurred.

Using the Webapp¶

You can also run the Pipeline using the Webapp. To do so, go to the Webapp in your browser and find the Pipeline you just posted in the Pipelines section.

Click on the VIEW button and you should see the Pipeline as a Flow.

Click on the UPLOAD button and upload an image. Now you can click on the RUN button and the Pipeline will be executed. When the Pipeline is finished, you can download the result by clicking on the DOWNLOAD button that will be enabled.

If the picture you provided had a face, the result should be blurred.