Load Caesar-stored annotations into the subject viewer outside drawing task context

[lcjohnso]

Package

lib-classifier

Problem Description

There are a few different situations where it would be beneficial to overlay externally-computed drawing annotations (e.g., ML predictions), typically in a passive, read-only way, outside of correct-a-machine drawing tool based editing workflows. Examples include:

• Viewing ML prediction and/or consensus annotations in Talk.
• Wanting to run simple "need correction or not" Yes/No question task workflow for ML predictions, but currently ML prediction annotations loaded from Caesar can't be displayed outside the drawing task context.

Describe the solution you'd like

Primary Feature

• For non-drawing workflows: use a workflow configuration key to indicate that Caesar should be queried for an annotation to be loaded into the subject viewer. Perhaps the same key envisioned in this comment or an additional, separate key for ready-only viz-only use case.
• For Talk (harder challenge): use a (private) subject metadata key-value pair to indicate what Caesar workflow can be queried to find an annotation that can be displayed.
  • NOTE: a challenge arises when we break the assumption that we are loading annotations stored as reductions on a specific Caesar workflow into a context other than the Classifier viewer for that specific workflow. Example: for viewing annotations in Talk, there isn't a default Workflow ID that points to where the annotations are stored in Caesar (e.g., as reductions on Workflow 999).
  • Possible Solution: adjust the default behavior from loading machineLearnt annotations into workflows and instead load them as project-level reductions in Caesar. This alleviates the current workflow-specific mapping of annotations, but still is not a fool-proof solution (i.e., subjects can be used across projects, so while the case is rare, it is not necessarily the case that looking up the subject's host project would be the correct place to fetch Caesar annotations).

Additional Features

• Add a toggle / checkbox on the viewer so that loaded annotations can be hidden/shown as desired.

Additional context

• See this Zooniverse Slack thread (access permissions required) for initial exchange clarifying needs and requirements based on initial trail us of correct-a-machine feature by the Science Scribbler: Synapse Safari project.
• This issue is thematically related to #6038 in that the proposed enhancement is a further extension of correct-a-machine capabilities.
• This issue could connect to Data Toolkit effort via link to consensus/aggregated results (in the form of image overlay) -- the capability for a new Talk landing page for subjects could feature a viewer that optionally showed consensus results from Caesar.

[eatyourgreens]

AFAIK, this isn’t documented but wiring a task up to request reductions is as simple as adding caesarKey to the task model.

https://github.com/zooniverse/front-end-monorepo/blob/5bcd1c865072035fe27ecce43acc58b69acf4e58/packages/lib-classifier/src/plugins/tasks/experimental/transcription/models/TranscriptionTask.js#L15

Once that’s done, the useCaesarReductions(task.caesarKey) hook should request reductions for any task.