Capture models - unorganised selectors

[stephenwf]

-- This proposal is a work in progress --

What is the problem

When a contributor wants to annotate a region of an image they have to decide where in the capture model the region should be associated. This is done by choosing a field or entity in a capture model and then choosing "define region". The region is then associated with the chosen field. This model has the following limitations

• Drawing a region requires a capture model to be created and available
• Regions cannot be transferred from one field to another
• If an external source wants to provide regions (e.g. OCR or external annotation tool) it has to create capture model fields - making it difficult for this to be generic and reusable.

These limitations have been worked around in various ways, however this relatively small change would give us a lot more flexibility and open up the ecosystem more.

What is the solution

We could introduce a new way to store these regions that is not directly attached to the capture model. This list of selectors could be populated by external tools (OCR or an annotation tool) in a step before filling out the capture model.

These unorganised selectors could then be presented when working filling in a capture model as an alternative to creating a new selector.

These unorganised selectors could be scoped outside of an individual project or specific to a project. This would allow these selectors to be reused.

Case study: Pre-segmentation

I want to run through a manifest and highlight all of the illustrations so that when I create a project those regions would be available to choose.

• I create a new "bucket" with an identifier to store my annotations in a manifest
• I want to open an annotation tool directly on the canvas page outside of a project
• When I create an annotation, it will be saved to the "bucket" I created
• Once complete, I want to tell a project to use the "bucket" I created for all fields in the capture model OR
• I want to tell a project to use the "bucket" I created for a single capture model field
• When a user crowdsources an image in my project they can choose one of the existing selectors instead of creating a new one

Case study: OCR generation

As a developer I want to be able to run some image recognition software to identify columns of text before a transcription project starts.

• I have created a project and have written some code to automatically generate bounding boxes on canvases
• My code needs to post these bounding boxes to an API endpoint (into a bucket)
• I then want to be able to use this bucket of selectors in my project
• I have the option before starting the project to pre-create fields in the capture model based on these regions
• When the user goes to annotate, they will see an empty field for each of the regions I generated
• The user can correct any mistakes in the regions

Case study: Separate Annotation tool

I want to create annotations using a different annotation tool and then associate those annotations to fields in a capture model in order to annotate more complex shapes and regions.

• I want to use a more advanced tool for highlighting regions on a canvas as part of the crowdsourcing process
• The output of this annotation tool creates regions that can be displayed, but not edited by the viewer
• When a user goes to contribute they will first be given my separate annotation tool to define regions
• Then a user will be able to fill out the fields in the capture model and choose the regions they created
• At any point the user can switch between the model and the annotation
• A user should be notified if a selector has not yet been associated with a capture model field

Note: It may be possible to map the output of the separate annotation tool directly to a capture model without the user having to fill out the capture model at all. This would require a limited capture model format and could be driven using profiles in the capture model itself.