Task Types

Below is a description of all possible task types. Depending on the task type, workers will be presented with different interfaces to label data.

Some task_types are only available with specific data_types, and may accept specific parameters.

classification

A simple classification task to associated items to classes. Supports single-choice ("class") and/or multi-choice ("tags").

parameters:

• classes: list of class options (single choice)
• tags: list of tag options (multi choices)

data_types : All

text

For transcription, text entry, sequence-to-sequence, etc. The annotator will have to fill a free text area.

parameters: None

data_types : All

bounding_box

An image annotation task where accepted regions are rectangular.

parameters:

• classes: list of region-level class options (single choice)
• tags: list of region-level class options (multi choices)
• global_classes: list of image-level class options (single choice)
• global_tags: list of image-level tags options (multi choices)

label format:

JSON object with the following properties:

• class: image-level class
• tags: image-level list of tags
• regions: list of objects with the following properties:
  • xmin, ymin, xmax, ymax: coordinates between 0 and 1 (as a percentage of width and height)
  • type: "box"
  • class: region-level class
  • tags: region-level list of tags

data_types: image

polygon

An image annotation task where accepted regions are polygons.

parameters:

• classes: list of region-level class options (single choice)
• tags: list of region-level class options (multi choices)
• global_classes: list of image-level class options (single choice)
• global_tags: list of image-level tags options (multi choices)

data_types: image

point

An image annotation task where accepted regions are points.

parameters:

• classes: list of region-level class options (single choice)
• tags: list of region-level class options (multi choices)
• global_classes: list of image-level class options (single choice)
• global_tags: list of image-level tags options (multi choices)

data_types: image

image_annotation

An image annotation task where accepted regions are of possibly different shapes.

parameters:

• region_types: the accepted region shapes (default to all)
• classes: list of region-level class options (single choice)
• tags: list of region-level class options (multi choices)
• global_classes: list of image-level class options (single choice)
• global_tags: list of image-level tags options (multi choices)

data_types: image

ner

Named Entity Recognition for textual data.

parameters:

• classes: list of class options (single choice)

data_types: text

audio_annotation

For audio data, regions annotations also referred to Part of Speech labelling (POS)

parameters:

• classes: list of class options (single choice)

data_types: audio

file_upload

The annotator will have to upload (a) file(s). Experimental support for small files only.

parameters:

• multiple: Boolean, whether multiple files are allowed
• files_accepted: list of file extensions allowed, e.g. [".jpg", ".png"]. For all image formats you can simply pass ["image"]

data_types: All

custom

Custom task based on a JSON Schema. JSON Schema is a standard format that makes it possible to define any data structure, and as a result generate automatically a corresponding HTML form for annotators to fill. With this task_type you can set up complex annotation tasks: nested attributes, lists, conditional fields, etc.

parameters:

• output_schema: a valid JSON Schema. (Documentation in progress). Take a look at this JSON format playground for ideas of what can be done.
• ui_schema: a JSON object describing the UI interface (Documentation in progress)

data_types: All