Creating & organising views
Concepts are the way to decompose and describe your visual universe. The term is actually a shortcut, as you should think more about visual concepts, i.e. concepts that have a visual unity of their own.
A View is a subset of a project. It is defined by logical conditions on inherited concepts and in turn defines its own set of concepts, thus giving rise to a hierarchical visual ontology.
A Root View corresponds to the first level of the hierarchy without any condition. A Root View then applies to all images of a project.
Parent and Child Views are linked together, the Child View being defined from conditions on the concepts of the Parent view. A Child View thus applies to a subset of the images of the Parent View.
To create a root view, you need to click on Add a root view if it's your first root view, or on Create a root view if you have already created one. You then need to choose a name for it.
A root view concerns all the images that have been added to the project.
To create a child view, there are two possibilities:
- you can click on Create a view at the bottom of a column, this will create a new child view with the same parent view as the other views of the column.
- you can click directly on the + button of a given view. This given view will automatically be the parent view of the view you create.
When you create a child view, you must specify conditions based on the concepts of the parent view to select the images that will be added to the view you are creating.
- when the parent view is a detection or tagging view, then you can use AND or OR conditions, and choose the concepts on which you want to filter.
The child view applies to images that have been annotated in the parent detection view with both snowboard and traffic light concepts..
You then need to select the type of view you want to create:
- Classification: each image is labelled with exactly one concept and all concepts are then exclusive.
- Tagging: each image is labelled positively or negatively for all concepts. A single image can then validate several concepts (or none).
- Detection: concepts are localised in the image with bounding boxes. The information is then much richer as you can get several instances of a same concept in the image.
You need to specify the concepts that you want to recognize. You can do so by clicking on New concept and by giving a name to your new concept. Don't worry, you can add more concepts and/or edit them after you have ended the project creation process.
You can also use existing concepts by importing a JSON file in which the concepts and images are defined. To do so, you must precise the field "tags" in the JSON. Check the JSON Upload page to know more about the JSON format accepted.
When entering a view, you access a screen where you can navigate through your Views and understand the hierarchy between all the Views.