Project Type

The project type indicates the annotations that are associated to each dataset. Four project types are supported: segmentation, landmarks, bounding boxes and classification.

The type of a project changes the tools that are available in the annotation tab, as well as the data that will be written during the export process.

A project may have multiple annotation types enabled at the same time, in which case the different annotations will be associated to the same dataset. During export, the user can select which annotation type will be exported.

With layers, multiple set of the same annotation type can be defined, with only one of them being active at a time. See the Layers page for more information.

Image Segmentation

This is the main and most featureful annotation type. In this type, a label map is associated to each dataset. The label map encodes one label per pixel of the dataset, and is intialized to the implicit “background” label. The label map can be modified in the Annotation tab using any of the tools. A full overview of the purpose of each tool is available in Annotation.

Note

Example if you want to segment livers and tumours in your images, create 2 labels: one for Liver, one for Tumours. You do not need to create a class for the “background” (i.e. the remaining pixels).

In this type of project, the Experiments tab is available, see Experiments for more information.

When exporting, the label map is exported together with the input data, and each label can be given a different output value.

Landmarks Annotation

In this type, multiple landmarks can be defined for each dataset. A landmark is a point within an image that represents a particular feature. Multiple types of landmarks can be defined to represent different types of features. Each type may contain multiple landmark instances.

During export, a label map is generated by giving a radius to each label, and a value depending on the landmark type.

Bounding Boxes Annotation

This type is similar to the Landmarks Annotation, except that the features are represented by bounding boxes.

Image Classification

In this type, data tags are used to classify each dataset, in addition to their usual filtering purpose. The Annotation tab is limited to visualizing the datasets.

Note

Example if you want to classify images in your database as healthy or not, create a “Healthy” tag and assign it to the relevant images after reviewing them in the annotation tab.

When exporting, a 1x1x1 label map is created for each dataset. The value of the pixel can be specified based on a combination of tags.