Appendix: How it differs from the old annotations

Basic Features

Data Source

Both the new and old annotation tools allow you to specify the ABEJA Platform data lake as the destination for importing tasks. The new annotation tool additionally allows you to import tasks by uploading a CSV which includes list of data sources accessible via HTTP.

Permission Management

The new annotation tool has three roles: Annotator, Annotation Manager, and Project Manager, and each user can have a different role. (Note that the Project Manager role is only available to users authenticated on the ABEJA Platform; users authenticated on the ABEJA Platform will automatically be in the Project Manager role.

Assignment Management Features

• Assignment management by project (work)

With the old annotation tool, projects created for users belonging to the same organization are automatically assigned. With the newer version, you can narrow down the target users and assign them to works within a project, making it possible to manage the scope of permission flexibly.

• Assignment Management by Task

Both the new and old annotation tools allow you to manage assignments per task. However, the old annotation tool only allows you to delete assignments to tasks, whereas the new annotation tool allows you to explicitly change a asignee for a task.

Notification Function.

With the new annotation tool, users associated with a target task will be notified of a status change when a status change is made, such as a result rejction. This helps annotators and annotation managers to clarify which tasks need to be checked to promote project efficiecy.

Project Duplication Feature.

Projects can only be duplicated in the old annotation tool. However, the new annotation tool also allows you to work with the same label definitions and data sources by creating additional work in the same data source within the project.

PreInference.

Both the old and new annotation tools support the ability to register pre-inference results via the API.

For details, see「pre-inference function」or「(old) pre-inference function」.

Export function.

JSON export

Both the old and new annotation tools support the JSON export function. However, the JSON format differs between the old and new annotation tools.

The JSON format of the old annotation tool

[
  {
    "id": 9999999,
    "project_id": 1111,
    "project_name": "example",
    "project_kind": "polygon",
    "task_id": 8888888,
    "skipped": false,
    "information": [],
    "worker_id": 1,
    "notes": null,
    "status": "notyet",
    "review_information": null,
    "created_at": "2020-01-01T10:00:00.000Z",
    "created_at": "2020-01-01T10:00:00.000Z",
    "task": {
      "id": 7777777,
      "priority": 1000,
      "status": "notyet",
      "notes": null,
      "created_at": "2020-01-01T10:00:00.000Z",
      "updated_at": "2020-01-01T10:00:00.000Z",
      "information": null,
      "metadata": [
        {
          "id": 6666666,
          "information": {
            "filename": "example.jpg",
            "timestamp": "2020-01-01T00:00:00+09:00",
            "x-abeja-sys-meta-organizationid": "1234567890123"
          },
          "created_at": "2020-01-01T00:00:00.000Z",
          "updated_at": "2020-01-01T00:00:00.000Z",
          "source": "20200101T000000-00000000-0000-0000-0000-000000000000",
          "channel_id": "1234567890987"
        }
      ]
    }
  }
]

New Annotation Tool JSON Format

[
  {
    "id": 7777777,
    "work_id": 10,
    "result": {
      "id": 9999999,
      "user_id": 1,
      "information": [],
      "worked_at": "2020-01-01T10:00:00.000Z"
    },
    "task_source": [
      {
        "source_uri": "datalake://1234567890987/20200101T000000-00000000-0000-0000-0000-000000000000",
        "filename": "example.jpg"
      }
    ]
  }
]

The new annotation tool changes or removes redundant fields that correspond to the following in the old annotation tool.

• status, skipped, result.status

  • With the new annotation tool, only reviewed tasks are output to the JSON results, so all included tasks can be considered as completed statuses. Therefore, the status field has been omitted.

• The project_id, project_name

  • Since the output JSON contains only the results of the same project, the project information in the fields of each task is omitted.

• task.metadata.channel_id, task.metadata.source

  • In the new annotation tool, the task_source.source_uri field is a URI that includes the channel ID and file ID.

• worker_id

  • In the new annotation tool, it is stored in the result.user_id field.

ABEJA Platform Dataset export.

Both the old and new annotation tools support exporting Dataset to the ABEJA Platform.

For more information, see Export results or (old) management screen description

Template types.

Supported templates are different between the new and old annotation tools. Currently, the old annotation tools support more templates than the new one.

For details, please check the list of supported templates for each tool.

• List of templates supported by the old annotation tools
• List of templates supported by the new annotation tool